Long-range time-of-flight scanning sensor based on …...Long-range time-of-flight scanning sensor...

11
Long-range time-of-flight scanning sensor based on high-speed time-correlated single-photon counting Aongus McCarthy, 1 Robert J. Collins, 1 Nils J. Krichel, 1 Verónica Fernández, 1,2 Andrew M. Wallace, 1 and Gerald S. Buller 1, * 1 School of Engineering and Physical Sciences, Heriot-Watt University, Riccarton, Edinburgh, UK, EH14 4AS 2 Now at Instituto de Física Aplicada, Consejo Superior de Investigaciones Científicas, Serrano 144, 28006 Madrid, Spain *Corresponding author: [email protected] Received 7 July 2009; revised 11 October 2009; accepted 12 October 2009; posted 13 October 2009 (Doc. ID 113822); published 4 November 2009 We describe a scanning time-of-flight system which uses the time-correlated single-photon counting tech- nique to produce three-dimensional depth images of distant, noncooperative surfaces when these targets are illuminated by a kHz to MHz repetition rate pulsed laser source. The data for the scene are acquired using a scanning optical system and an individual single-photon detector. Depth images have been suc- cessfully acquired with centimeter xyz resolution, in daylight conditions, for low-signature targets in field trials at distances of up to 325 m using an output illumination with an average optical power of less than 50 μW. © 2009 Optical Society of America OCIS codes: 280.3400, 120.3930, 040.3780, 150.6910. 1. Introduction In recent years, application areas for three- dimensional imaging have emerged in several fields, including manufacturing [1], defense [2], and geo- sciences [3]. Several groups have acquired depth in- formation and constructed depth images using varia- tions of the time-correlated single-photon counting technique (TCSPC) [49]. The TCSPC approach to laser radar affords a number of potential advantages over non-photon-counting approaches. These poten- tial advantages include improved depth resolution [10] over linear multiplication detector-based sys- tems [11]. The improved depth resolution means that TCSPC can be effectively utilized in the identifica- tion of distributed targets where the scattering sur- faces are closely separated [12]. The advantage of shot-noise-limited detection of single-photon events means that the system can be used even when there is an average of less than one photon return event per pulse, which is critical in applications involving long-distance ranging [5]. At kilometer ranges, the low photon return rates permit the use of high repe- tition rate, compact, low-power laser diodes, since it is not necessary to record a return on every outgoing laser pulse. The TCSPC technique does have several potential disadvantages, including the following: limited spectral range of practical single-photon de- tectors [13]; detector dead time where the detection system is shut down to reset after the recording of an event [14]; and issues regarding solar background events providing potential false alarms [15]. None- theless, there is growing interest in depth imaging using TCPSC, especially as photon-counting detector and acquisition technology continues to improve, per- mitting faster data acquisition at longer range. The general approach of photon-counting ranging has been to direct a pulsed laser toward a noncoo- perative target (i.e., an object or scene that does not lend itself to reflecting the transmitted laser pulses back toward the source) with the scattered photon re- turn being recorded by a photon-counting detector. Timing information from the recorded photon re- turns is used to determine the round-trip time, which subsequently is employed to calculate the distance to 0003-6935/09/326241-11$15.00/0 © 2009 Optical Society of America 10 November 2009 / Vol. 48, No. 32 / APPLIED OPTICS 6241

Transcript of Long-range time-of-flight scanning sensor based on …...Long-range time-of-flight scanning sensor...

Page 1: Long-range time-of-flight scanning sensor based on …...Long-range time-of-flight scanning sensor based on high-speed time-correlated single-photon counting Aongus McCarthy,1 Robert

Long-range time-of-flight scanning sensor based onhigh-speed time-correlated single-photon counting

Aongus McCarthy,1 Robert J. Collins,1 Nils J. Krichel,1 Verónica Fernández,1,2

Andrew M. Wallace,1 and Gerald S. Buller1,*1School of Engineering and Physical Sciences, Heriot-Watt University, Riccarton, Edinburgh, UK, EH14 4AS

2Now at Instituto de Física Aplicada, Consejo Superior de Investigaciones Científicas,Serrano 144, 28006 Madrid, Spain

*Corresponding author: [email protected]

Received 7 July 2009; revised 11 October 2009; accepted 12 October 2009;posted 13 October 2009 (Doc. ID 113822); published 4 November 2009

We describe a scanning time-of-flight systemwhich uses the time-correlated single-photon counting tech-nique to produce three-dimensional depth images of distant, noncooperative surfaces when these targetsare illuminated by a kHz to MHz repetition rate pulsed laser source. The data for the scene are acquiredusing a scanning optical system and an individual single-photon detector. Depth images have been suc-cessfully acquired with centimeter xyz resolution, in daylight conditions, for low-signature targets in fieldtrials at distances of up to 325m using an output illumination with an average optical power of less than50 μW. © 2009 Optical Society of America

OCIS codes: 280.3400, 120.3930, 040.3780, 150.6910.

1. Introduction

In recent years, application areas for three-dimensional imaging have emerged in several fields,including manufacturing [1], defense [2], and geo-sciences [3]. Several groups have acquired depth in-formation and constructed depth images using varia-tions of the time-correlated single-photon countingtechnique (TCSPC) [4–9]. The TCSPC approach tolaser radar affords a number of potential advantagesover non-photon-counting approaches. These poten-tial advantages include improved depth resolution[10] over linear multiplication detector-based sys-tems [11]. The improved depth resolution means thatTCSPC can be effectively utilized in the identifica-tion of distributed targets where the scattering sur-faces are closely separated [12]. The advantage ofshot-noise-limited detection of single-photon eventsmeans that the system can be used even when thereis an average of less than one photon return eventper pulse, which is critical in applications involving

long-distance ranging [5]. At kilometer ranges, thelow photon return rates permit the use of high repe-tition rate, compact, low-power laser diodes, since itis not necessary to record a return on every outgoinglaser pulse. The TCSPC technique does have severalpotential disadvantages, including the following:limited spectral range of practical single-photon de-tectors [13]; detector dead time where the detectionsystem is shut down to reset after the recording of anevent [14]; and issues regarding solar backgroundevents providing potential false alarms [15]. None-theless, there is growing interest in depth imagingusing TCPSC, especially as photon-counting detectorand acquisition technology continues to improve, per-mitting faster data acquisition at longer range.

The general approach of photon-counting ranginghas been to direct a pulsed laser toward a noncoo-perative target (i.e., an object or scene that does notlend itself to reflecting the transmitted laser pulsesback toward the source) with the scattered photon re-turn being recorded by a photon-counting detector.Timing information from the recorded photon re-turns is used to determine the round-trip time, whichsubsequently is employed to calculate the distance to

0003-6935/09/326241-11$15.00/0© 2009 Optical Society of America

10 November 2009 / Vol. 48, No. 32 / APPLIED OPTICS 6241

Page 2: Long-range time-of-flight scanning sensor based on …...Long-range time-of-flight scanning sensor based on high-speed time-correlated single-photon counting Aongus McCarthy,1 Robert

the target. Each transmitted pulse and photon re-turn constitutes an individual measurement of thedistance to the target. Statistical analysis of manyof these independent measurements (typically 103–106) can yield a time-of-flight resolution that is betterthan the overall timing jitter of the system. In priorwork [8,16,17], we constructed a system based onTCSPC for performing depth image measurementsin indoor environments under typically low ambientlight level conditions at standoff distances of around10m. This used a large pan-and-tilt head mechanismto scan a compact optical system containing an opti-mized silicon single-photon avalanche diode (SPAD)detector. In parallel, we developed a kilometer rangesystem, based on a commercially available 200mmdiameter aperture Schmidt–Cassegrain telescopefor single point use in target identification underdaylight conditions [12].Time-of-flight imaging systems using arrays of sin-

gle-photon detectors have been reported, for exam-ple, by Priedhorsky et al. [9], who proposed asystem using a microchannel plate with cross-chan-nel delay line to provide time-of-flight informationwith x − y spatial information. This proposal was de-monstrated later by Ho et al. [6] in both indoor andoutdoor environments over a range of 50m. More re-cently, a group at the Massachusetts Institute ofTechnology Lincoln Labs demonstrated photon-counting time-of-flight using 4 × 4 arrays of detectors[4] and scanning mirrors with measurements atranges of 60m shown for target objects in daylight.Later, the group made time-of-flight measurementsusing a 32 × 32 photon-counting detector array [7]and a rotating Risley prism beam steering arrange-ment to form depth images at ranges of 150m. In [5],Degnan describes a low dead-time photon-countingsystem used as a time-of-flight microaltimeter cap-able of measuring multiple return events from asingle laser pulse, using a microchannel plate photo-multiplier approach. In this case, the source was op-erating at kHz repetition rates and the systemexhibited measurements at altitudes of several km.While arrayed detectors can provide advantages in

terms of parallelism, there can be issues with crosstalk and fill factor [18], in particular. Here wepresent a scanning time-of-flight depth sensor thatuses a commercially available individual, ungatedsingle-photon counting detector module. The high-performance single-photon detector module is fi-ber-coupled to a compact optical transceiver headthat contains a pulsed laser diode and a pair of gal-vanometer scanning mirrors. These mirrors are usedto scan both the outgoing laser beam and the col-lected return photons—meaning that a single opti-mized detector can be used to obtain a completethree-dimensional image. The single-mode optical fi-ber delivery method means the system can be easilyreconfigured for use with alternative single-photondetector technologies and the fiber also acts as a con-venient and effective spatial filter for backgroundlight. Unlike many of the other systems described,

this system is configured for low energy opticalpulses (typically <30pJ or fewer than ∼130 × 106

photons per pulse) and this means that multiplephoton returns from a single output pulse are im-probable. This low probability helps to minimizethe number of collected return photons that go unde-tected—a single triggering detector is used and, oncea photon event has been recorded, the subsequent de-tector dead time (currently ∼100ns) means that it isunable to register any additional photons arriving inthis period. We utilize low energy optical pulses(∼pJ) for this work in conjunction with a potentiallyhigh repetition rate laser source (tens of MHz) to re-duce the data acquisition time. The inherent timegating and high sensitivity of the TCSPC approachand effective spatial and spectral filtering permitdaytime and nighttime operation at eye-safe laserpower levels. The system has been used in daylightconditions to construct depth images with centimeterspatial and depth resolution of noncooperative tar-gets at a range of 325m using μW average outputpower levels. This range distance does not representa limitation of the system, but instead was dictatedby available target range facilities during these fieldtrials. The two scanning mirrors are common to boththe transmit and the receive channels and enable thescene to be scanned much faster compared with therelatively slow data acquisition process when usingthe pan-and-tilt head of our previous system [17].The coaxial optical arrangement means that thetransmitted beam is directed selectively, only atthe field point being ranged, leading to the most effi-cient use of the low-power source. In the measure-ments shown below, the average power of theoutput illumination used was less than 50 μW.

The system described here is shown schematicallyin Fig. 1 and a summary of the main system param-eters is listed in Table 1. The system uses a pico-second-pulsed semiconductor diode laser whoseoutput is scanned across the scene of interest by agalvanometer mirror pair. The laser typically oper-ates at kHz toMHz repetition rates, with the averagetransmitted optical power being between 1 μW and1mW. The scattered optical return from the distanttarget is collected by the same optical system from

Fig. 1. Schematic diagram indicating the principal components ofthe scanning system. The scene typically is at a range of hundredsof meters to several km. Electrical paths are denoted by solid linesand optical paths by dashed lines. Si-SPAD is a silicon single-photon avalanche diode.

6242 APPLIED OPTICS / Vol. 48, No. 32 / 10 November 2009

Page 3: Long-range time-of-flight scanning sensor based on …...Long-range time-of-flight scanning sensor based on high-speed time-correlated single-photon counting Aongus McCarthy,1 Robert

which the laser pulse is emitted and redirected usingpolarization optics to a single-mode optical fiber de-tection channel. This fiber is connected to a single-photon detector module and the output signal fromthe detector is recorded as time-tagged photon eventsthat are continuously streamed to the control compu-ter. The control software reconciles this timing in-formation with field position to calculate depthsat individual target positions and, thus, produce adepth image across the scanned field.

2. Optical Design and Optomechanics

A schematic diagram of the optical system is shownin Fig. 2. An interchangeable commercially availablesingle lens reflex (SLR) camera lens is used as thesystem objective. This objective is used to directthe outgoing laser to the target and also to efficientlycollect the scattered photons returned from thetarget. For the experiments detailed in this paper,the objective was a 200mm effective focal length,

Table 1. Summary of the Scanning Sensor System Parameters

Parameter Value / Comment

Transmit/receive alignment Mono-static, i.e., transmit and receive channels are coaxialLaser wavelength λ ∼842nmLaser type Pulsed semiconductor diodeLaser pulse width ∼90ps, full width at half-maximumOptical output power POut <50 μW averageDetector Si-SPAD, Single-Photon Counting Module (Perkin Elmer),

Active area diameter 180 μm, thick junction, ∼400ps jitter, fiber-coupled(∼5 μm diameter core, armored fiber)

Beam scanning mechanism Galvanometer mirror pair (x − y) (common to both the transmit and receive channels)System objective 200mm focal length, f =2:8 (Canon EF camera lens)Receive aperture diameter 72mmSystem f =# f =4Field of view Total available for scanning: ∼55mrad (17m diameter at 325m)

Single point: ∼70 μrad (22mm diameter at 325m). The control software allowsthe user to set the xy extent of the scan within the available field of view,and the number of “pixels” in x and y.

Range Scans carried out at, but not limited to, ranges from 5 to 330m.System resolution at 325m Spatial (xy): ∼20mm

Depth (range): As low as 20mm. The depth resolutionis dependent on a number of factors but primarily the number of detectedreturn photons (the longer the acquisition time the better the resolution).

Fig. 2. Schematic of the optical layout. The pulsed output of the laser diode is incident on a pair of steering mirrors (TM1, TM2), passesthrough a half-wave plate (HWP), a polarizing beam splitter (PBS), two galvanometer scanning mirrors (SM1, SM2), relay optics and exitsthe transceiver head through the camera lens objective. A small percentage of the incoming light is reflected into a monitoring channelwhere an edge filter (EF) is used to block the laser transmission. The return path goes through the same scanning and relay optics as thelaser beam, passes through the beam splitter, onto a pair of steering mirrors (RM1, RM2) and is spectrally filtered by a bandpass filter(BPF) before being coupled into an ∼5 μm core diameter single-mode fiber (SMF) by a fiber collimation package. The fiber is connected tothe single photon avalanche diode (SPAD) detector module.

10 November 2009 / Vol. 48, No. 32 / APPLIED OPTICS 6243

Page 4: Long-range time-of-flight scanning sensor based on …...Long-range time-of-flight scanning sensor based on high-speed time-correlated single-photon counting Aongus McCarthy,1 Robert

nine-element Canon camera lens with an f numberof f =2:8.The required part of the scene is relayed to the

optical fiber core (which is connected to the single-photon detector) via the imaging optics and a pair ofcomputer-controlled galvanometer scanning mirrors.The mirrors SM1 and SM2 as shown in Fig. 2 and therequired lenses are arranged in a spatially separatedtelecentric relay configuration. The scanning mirrorsare placed at conjugate planes of the system and, as aresult, the point at which the transmit and receivebeams are incident on the mirrors remains station-ary as they are scanned. Three identical five-ele-ment, 30mm focal length lenses [19], operating atinfinite conjugates are used for the relay optics. Thisfive-element lens design has a flat field, diffraction-limited performance at 850nm and is corrected forf sinΘ distortion. An 11mm focal length fiber colli-mation package is used to couple the light into asmall (<10 μm) diameter optical fiber core connectedto the detector. This arrangement provides an effec-tive means of spatially filtering the backgroundillumination, such as solar background and off-axis light.The maximum field of view that can be scanned by

this sensor is primarily set by the field of view of theimage presented to the internal system optics by theobjective lens, and the field of view of the 30mm focallength relay lenses, the latter being the limiting fac-tor. The 200mm focal length objective and 30mm fo-cal length relay lens combination used for the workdescribed in this paper has a maximum full field ofview of 55mrad. The spatial resolution achieved de-pends on a number of factors including the focallengths of the lenses, the aberrations of the optics,and, in particular, the size of the fiber core coupledto the detector. When the system was aligned and fo-cused on the scene at 325m, the aperture of the re-ceive channel when using a 5 μm diameter core fiberwas measured to be approximately 25mm in dia-meter with the transmit channel laser beam dia-meter being slightly larger. This beam diameterequates to a spatial resolution of approximately70mm at a 1km range and corresponds to the mini-mum angular step resolution of the galvanometermirrors used in the measurements shown below.The sensor head consists of a custom-built optome-

chanical assembly that uses a slotted-baseplateapproach for mounting the majority of the opticalcomponents [20]. A 25mm thick plate of aluminum,with an appropriate network of slots of a fixed widthmachined into its surface, is used as the baseplate forthis sensor. The optical components and devices aremounted in 35mm diameter magnetic steel barrels,which in turn are placed in the baseplate slots andheld in position with magnets. The combination ofslot width and barrel diameter defines an optic axiswhich, in this case, is 14mm above the top surface ofthe baseplate. This optomechanical system providesa convenient and robust semi-kinematic mount thatcombines good long-term mechanical stability under

typical laboratory conditions, with the possibility ofreconfiguration and ease of alignment. The objectivelens is attached to the baseplate using a Canon lensfitting adaptor plate—thus enabling any lens with acompatible fitting to be used with the system. Thealuminum baseplate, sidewalls and lid that makeup the enclosure were surface finished with an∼10 μm thick black anodized coating to protect thealuminum while handling as well as helping to mini-mize internal stray light levels. The completedsensor head is shown in Fig. 3 and measures∼275mm × 275mm × 170mm.

The transmit and receive channels of the systemwere aligned to be coaxial so that the image of thereceiver collection fiber core was concentric with thetransmitted laser beam, irrespective of the range.This was achieved by monitoring the near-fieldand far-field alignment of the transmit and receivechannels while adjustments were made using a pair

Fig. 3. (Color online) Transceiver head assembly: (a) Three-di-mensional image produced from the CAD model of the systemshowing the enclosure (∼275 × 275 × 170mm) and the slotted-baseplate optomechanics. The two galvanometer servo-control cir-cuit boards are housed on the underside of the slotted baseplate.(b) Photograph of the assembled transceiver head.

6244 APPLIED OPTICS / Vol. 48, No. 32 / 10 November 2009

Page 5: Long-range time-of-flight scanning sensor based on …...Long-range time-of-flight scanning sensor based on high-speed time-correlated single-photon counting Aongus McCarthy,1 Robert

of adjustable steering mirrors in both of the chan-nels. The sensor head maintained its optical align-ment over several months of operation in the fieldlaboratory environment where the temperature var-ied between approximately 10 °C and 25 °C.

3. Control Electronics and Data Acquisition

The master clock for the system used when acquiringthe results shown in this paper was a commerciallyavailable pulse pattern generator, which served tosynchronize and gate the TCSPC electronics andclock the laser pulses over a frequency range of a fewkHz to 2MHz. The laser was an ∼90 picosecondpulse-width laser diode operating at a central wave-length of ∼842nm and the output power was con-trolled by a variable current driving unit.The output laser pulses and returned photons

were scanned using a pair of galvanometer mirrors,one providing adjustment of the vertical (y) positionand one providing adjustment of the horizontal (x)position. Each of the mirrors was driven by a ser-vo-control circuit board and the position commandinput signals for these boards was set by the systemcontrol software through a computer-based digital-to-analog converter card.Data acquisition was performed using a PicoHarp

300 TCSPC module, manufactured by PicoQuantGmbH. Photons returned from the target and suc-cessfully collected by the objective lens were mea-sured using a fiber-coupled silicon single-photonavalanche diode (Si-SPAD). In this case a free-run-ning commercially available (Perkin Elmer) thickjunction silicon detector with an active area diameterof 180 μmwas used, although other fiber-coupled sin-gle-photon detectors can be connected for futuremea-surements. The optical head was connected to the Si-SPAD using an ∼5 μm diameter core optical fiberwith an angle polished connector (APC) to minimizeoptical backreflections, which would appear as ghostsecondary peaks in the target return histogram (seeSection 4 ).

4. Software

A. Hardware Control

The system was controlled by a desktop personalcomputer (PC) running custom software developedin house. The current iteration of the software wasdesigned for maximum flexibility in testing and de-velopment and was not optimized for speed of opera-tion or processing. The TCSPC module was operatedin time-tagged mode so that the times of events oneach of the two inputs were recorded independently,but relative to a common macrotime. One input wasthe synchronization signal from the master clock in-dicating the time at which a laser pulse was emitted,the other was the signal from the Si-SPAD, indicat-ing the time of a photon return.The camera lens that is used as the system objec-

tive introduced a backreflection into the returnphoton path, which resulted in a system return peak

in the histogram containing a high number of inte-grated counts. The TCSPC electronics had a maxi-mum count rate of approximately four million eventsper second across both input channels. Photons de-tected from the backreflection cause a high percen-tage of the overall photon return rate and do notcontribute to depth profile measurements. TheTCSPC electronics have an input that suspends ac-quisition of photon events when the transistor–tran-sistor logic (TTL) voltage applied to it is in a low logiclevel. A signal from the master clock was used to sus-pend registration of photon events during the ex-pected period of the internal optical backreflection,as shown in Fig. 4. This reduced the overall recordedcount per second rate at the TCSPC electronics andpermitted the use of higher laser pulse repetition fre-quencies. The relative time and duration of the elec-trical gate could be adjusted as required by theexperimental conditions.

B. Data Acquisition and Analysis

To acquire a depth profile, the galvanometer mirrorssteer the laser beam in a raster pattern over the spe-cified field of view in a move–stop–move motion. Thecontrol software allows the user to set the xy extent ofthe scan within the available field of view, the num-ber of pixels in x and y, and the pixel dwell time. TheTCSPC electronics commences the acquisition ofphoton return events at the beginning of a frame andcontinuously streams photon return data until theend of the frame. To mark the beginning of the frameand the end of each line, a low resolution (∼10nsFWHM timing jitter) marker is inserted into theraw time stream data. The analysis software subdi-vides the collected time stream data at each of thesemarkers to produce a fixed number of lines, and thenfurther subdivides each line into the appropriatenumber of individual pixels.

Fig. 4. Example histograms taken on a retroreflective referencematerial. Both measurements were taken with a 1 s acquisitiontime and feature comparable target return peaks around 165ns.The high repetition rate of the system results in multiple photonpulses being in transit at any given point and causes distant targetreturns to be assigned to short flight times. The lower plot demon-strates the hardware gating functionality: reducing the duty cycleto 80% removed significant photon returns caused by internal op-tical reflections of the system.

10 November 2009 / Vol. 48, No. 32 / APPLIED OPTICS 6245

Page 6: Long-range time-of-flight scanning sensor based on …...Long-range time-of-flight scanning sensor based on high-speed time-correlated single-photon counting Aongus McCarthy,1 Robert

The two independent streams of high-resolutionraw timing data were used to construct a histogramof counts for each pixel. Each time stamp correspond-ing to a transmitted laser pulse was used as a timingreference point. Between each transmitted laserpulse, the time stamps corresponding to detectorevents were added to a histogram with total durationequal to the laser pulse repetition period.Each individual detector event, some of which cor-

respond to photons returned from the target, is anindependent measurement of the distance. In anideal time-of-flight system, only one single-photonreturned from the target would be sufficient to makea measurement of the distance to the target. How-ever, uncertainty is introduced into the distancemea-surement from several factors, including timingjitter (from the master clock, the laser driver, the Si-SPAD and the TCSPC electronics) and spuriouscounts arising from detector dark noise and a contri-bution from other light sources, most typically that ofthe solar background. Hence, it is necessary to utilizemultiple counts, corresponding to many transmittedpulses. The effect of timing jitter will be evident as abroadening of the peak in the histogram and the ef-fect of spurious counts is an increase in the overallbackground counts. A cross-correlation algorithmwas used to identify the peak corresponding to thetarget return by correlating it with a known detectorresponse [21]. By measuring the distance to the tar-get object at each pixel, a depth can be estimated. Toreduce the amount of data processing, in cases wherethe approximate range to the target was known, itwas possible to restrict the portion of the histogramthat was checked for a correlation to the depth regionnear the expected range.Many rangefinders use a low pulse repetition fre-

quency to ensure that only one pulse is in transit atone time [3] to avoid ambiguity in the absolute rangeby aliasing from adjacent optical pulses. In the mea-surements presented in this paper, we used frequen-cies of up to 2MHz, since depth profiling could beachieved without the concern of range ambiguity—the approximate target depth being unambiguouslydetermined via low repetition frequency measure-ments. It is possible to increase the pulse repetitionfrequency provided that the period between succes-sive laser pulses exceeds the return time of flight be-

tween the points that define the maximum surfaceseparation within the depth image. Without havinga priori knowledge of the scene being scanned, theimplementation of the system as described herewould require that the object of interest is in isola-tion and has a limited range of depth. In these mea-surements, the maximum pulse repetition frequencywas limited by the first generation of the custom con-trol software but later software versions have sincepermitted much higher laser repetition rates. Tech-niques based on the use of a pseudorandom modu-lated laser output have been used to overcome theproblem of range ambiguity in lidar systems, e.g.,[22,23]. More recently, a similar approach [24] hasbeen used for the complete avoidance of range ambi-guity at high laser frequencies in photon-countingsystems, although this technique has not been imple-mented in the results presented here. The approachdescribed in [24] is fully compatible with the low-power semiconductor laser diode sources used in thiswork.

5. Experimental Results

The current system presented in this paper was de-veloped as a generic test bed and used many multi-function devices with additional features surplus tothose required for a fixed, known application scenar-io. The system was capable of operation at frequen-cies in the range of a few kHz up to 2MHz andmeasurements were performed using several ofthese frequencies over a range of different ambientconditions. Table 2 summarizes the key scan para-meters for two representative depth profile measure-ments that were performed with the system, whileFigs. 5 and 6 show the depth profiles obtained.

The solar background count rate depended both onweather conditions and on the angle of the Sun rela-tive to the scanning head and was typically in therange of 3000 to 15; 000 counts=s when observed indaylight conditions. These solar background mea-surements were taken with the system pointing inthe direction of a reference target to make compara-tive assessments. The reference target was located ata range of 325m and was a large sheet of white retro-reflective material of a type similar to that used forBritish car license plates. The transceiver sensorhead was situated in a rooftop laboratory and

Table 2. Information on Two Representative Depth Profile Measurements

Object Imagea Conditions Range (m)Scan Dimensions

(width × height ðmÞ)

Resolution ofScanned Image

(pixels)

Field ofview b

(mrad)

PulseRepetitionFrequency

(kHz)

ApproximatePowerc

(μW)

PixelDwell

Time (s)

Car Fig. 5 Outdoorsat dusk

330 4:0 × 1:2 200 × 64 12 350 10 1.0

Mannequin Fig. 6 Outdoorson sunny day

325 0:8 × 2:0 32 × 128 6 2000 40 1.0

aFigures 5 and 6 show the depth profiles obtained for the measurements described in this table.bThe field of view column lists the maximum full angle.cTransmitted powers were measured outside the camera lens after the depth profiles had been obtained.

6246 APPLIED OPTICS / Vol. 48, No. 32 / 10 November 2009

Page 7: Long-range time-of-flight scanning sensor based on …...Long-range time-of-flight scanning sensor based on high-speed time-correlated single-photon counting Aongus McCarthy,1 Robert

pointed in a southwesterly direction when aligned onthe target site. This site was approximately 1° lowerin elevation than the rooftop laboratory. As men-tioned previously, target distances were limited toranges of around 325mdue to the available field trialfacility and do not represent range limitations ofthe system, as will be examined in more detail inSection 6.These results clearly demonstrate photon-count-

ing depth imaging on low-signature targets in a vari-ety of daylight and weather conditions. However, theresults were obtained using long overall acquisitiontimes, typically minutes, due to the relatively low la-ser repetition rates used in this system comparedwith our previous laboratory-based studies [17].Also, there was some degree of loss in the systemwhich led to the overall optical energy per pixel usedin these measurements being larger than necessary.The required energy per pixel (i.e., >1 μJ) was simi-lar to that used in depth imaging using linear multi-plication detectors [11], albeit our results were at

substantially longer range. Both these performanceissues are discussed in Section 6 and predictions ofpossible future performance are given based on theseexperimental results.

6. Modeling Predictions

The current working system presented in this paperprovided a test bed by which the performance of fu-ture configurations of the system could be predicted.For example, such analysis provides a means to ex-amine the trade-offs between data acquisition time,optical power, and target range under different oper-ating conditions. The return signal of the target canbe characterized in terms of signal-to-noise ratio(SNR). The SNR of a measurement can be definedas in [10]

SNR ¼ npffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffinp þ nb

p ; ð1Þ

where np is the number of counts in the highest chan-nel of the return peak in the time-of-flight histogram

Fig. 5. (Color online) 200 × 64 pixel image of a Peugeot 307 car taken at a distance of 330m in dusk lighting conditions. (a) Photograph ofthe car in situ on the target range. The dashed box indicates the approximate boundaries of the depth-profile measurement. (b) Depth/intensity image of the car generated from the processed depth information. The colors are defined by the intensity returns from each pixel.

10 November 2009 / Vol. 48, No. 32 / APPLIED OPTICS 6247

Page 8: Long-range time-of-flight scanning sensor based on …...Long-range time-of-flight scanning sensor based on high-speed time-correlated single-photon counting Aongus McCarthy,1 Robert

and nb is the average number of counts per channelin the background. As the background level is re-duced, the SNR is dominated by photon statisticsof the peak. Such a SNR approach is distinguishedfrom more basic approaches like analysis of thesignal-to-background ratio (SBR), which cannot ac-count for the return histogram improving with longerintegration times.The average background level can be estimated for

the employed repetition rate f Rep and binning sizetBin, based on averaged total count rates CBG for dif-ferent environmental conditions with the illuminat-ing laser switched off,

nb ¼ tAcqCBGf ReptBin; ð2Þ

where tAcq is the acquisition time per pixel.The number of illumination photons of wavelength

λ leaving the transceiver is readily determined basedon the average output power POut measured in frontof the unit. A theoretical model of the system can becreated based on this information and taking into ac-count all subsequent photon attenuation stages. Theone-way distance r of the scanned object is of rele-vance both to the atmospheric attenuation en-countered during the round trip and due to thedivergence of the illumination beam. The extinction

coefficient αMod for the negative exponential relationto distance caused by atmospheric attenuation wassimulated for a number of different weather and en-vironmental conditions using the MODTRAN soft-ware package [25]. An attenuation factor TMat isincluded to consider the significant photon loss dueto absorption and reflection of light by the target ob-ject that is not aimed back toward the scanning sys-tem. TTrans compensates for further attenuationinside the transceiver and caused by the limited de-tection efficiency of the single-photon detector. TLenscan be used to factor in the effect of changing the sys-tem’s camera objective. Ultimately, the detector re-sponse coefficient (DRC) converts the overallnumber of returned and detected photons into thenumber of photons in the maximum bin of the returnhistogram. Its value is constant for different peakheights but varies when either the binning size orthe detector type is changed.

All the factors mentioned are combined to deter-mine np:

nP ¼ tAcqPOutλhc

e−αMod×2r

2r2TLensTTransTMat

DRC; ð3Þ

where h is Planck’s constant and c is the speed oflight in vacuum. This allows one to calculate the

Fig. 6. (Color online) 32 × 128 pixel image of a life-sized mannequin taken at a distance of 325m in daylight conditions. (a) Photograph ofthe 1:8m tall mannequin in the scan position. (b) and (c) Three-dimensional plot of the processed depth information—the curvature of the600mm diameter concrete pillar behind the mannequin is clearly evident in (b).

6248 APPLIED OPTICS / Vol. 48, No. 32 / 10 November 2009

Page 9: Long-range time-of-flight scanning sensor based on …...Long-range time-of-flight scanning sensor based on high-speed time-correlated single-photon counting Aongus McCarthy,1 Robert

achievable SNR in conjunction with Eqs. (1) and (2).Values for the attenuation inherent to the systemwere determined based on power measurements atdifferent points within the transceiver. Coefficientsfor different target materials at different incidenceangles could be obtained by performing single-pixelscans on a range of materials with all other influen-cing quantities determined beforehand.Ultimately, the system is limited by the employed

algorithm’s ability to reliably detect the instrumen-tal response within a given histogram. In this paper,linear cross-correlation techniques were employed,although other approaches such as maximum likeli-hood estimation [26] and Markov chain algorithms[27] have previously been used successfully for thedetection of low-signature target returns in highbackground levels in photon-counting time of flight.Such algorithms [27] effectively permit reducedphoton return levels to be used for depth imaging,albeit at the expense of increased processing times.Reversible jump Markov chain algorithms allowfor the dynamic adaption of the number of degreesof freedom of the system, effectively allowing for ana-lysis of target areas with an unknown number ofpeak responses [28]. If the procedural method of peakidentification within any of these algorithms em-ploys either a mathematical model of the expectedsystem target response or photon return data takenat relatively long integration times as a reference,scintillation effects within analyzed data of shorteracquisition times can cause shape discrepancies det-rimental to the algorithm’s capability to producereliable depth estimates. These scintillations predo-minantly originate from irradiance and return inten-sity fluctuations due to atmospheric turbulence [29].A preprocessing step to minimize scintillations in re-turn histograms can be realized with a number ofapproaches, e.g., implementation of a nonlinear med-ian- or low-pass filter. Additionally, these algorithmscan facilitate reliable distance determination in tur-bid operational environments, such as fog, which canbe treated as an intermediate distributed target ob-ject, causing a high number of dispersed, low-inten-sity photon returns.To determine the minimum SNR value at which

the peak finder reliably locks on to a target returnfor a given binning size, the single-pixel scans men-tioned above were analyzed and the returned targetrange checked for consistency. Once a preliminaryminimum SNR value was estimated, additionalscans in the suspected regime were performed. Thevalidity of these experiments was verified with addi-tional scans in which the target return was kept con-stant against a varying background light level,achieved by an external, continuous light source withvariable brightness at close range.An example for the performance of the system in

terms of SNR against target range for three targettypes—aluminum, brass, and foliage—is shown inFig. 7 under the conditions of a 1ms acquisition timeand a transmit output power of 1mW. These values

are chosen to permit the acquisition of a depth imageof 32 × 32 pixels in approximately 1 s, while main-taining eye-safe operation. Also shown in the graphis the minimum SNR detectable using the cross-correlation algorithm. One can see from the graphthat, depending on the target material, the sensorwill work under these conditions at ranges of lessthan 1km.

For longer ranges, the sensor can operate at longerdwell times. However, use of the model does providethe opportunity to examine how system design canaffect potential range performance. For example,the sensor design contains three relay lenses thatprovide a source of backreflections and attenuationin the common transmit and receive channel. These

Fig. 7. (Color online) Prediction of SNR versus range for first gen-eration sensor for three different target surfaces at an acquisitiontime of 1ms. Themodel is based on experimental measurements ofphoton returns. The prediction is for a 40MHz pulsed laser withaverage system output power level of 1mW at 842nmwavelength.The model assumes overcast daylight conditions. The red line in-dicates the minimum SNR required for depth measurement usingelementary cross-correlation analysis. The model uses MODTRANestimates for atmospheric transmission in an urban environment.

Fig. 8. (Color online) Prediction of maximum detectable range forhigh- and low-signature targets versus acquisition time, showingpredictions for the first generation system and the likely perfor-mance of an improved system using a wavelength of 842nm, a la-ser repetition frequency of 40MHz, and an average optical powerof 1mW. Assumptions are based on a reduction of internal systemloss by 7dB, doubling the lens aperture diameter, and adaption ofthe transmitted spot size. The simulated laser output power, atmo-spheric conditions, and data analysis method are the same as inFig. 7.

10 November 2009 / Vol. 48, No. 32 / APPLIED OPTICS 6249

Page 10: Long-range time-of-flight scanning sensor based on …...Long-range time-of-flight scanning sensor based on high-speed time-correlated single-photon counting Aongus McCarthy,1 Robert

losses amount to more than 13 dB in the receivechannel between the camera lens and the optical fi-ber connected to the detector. While some of theselosses are unavoidable (for example, a 3dB lossdue to filtering of unwanted polarization states atthe polarizing beam splitter), there is considerablescope to reduce them in this first generation proto-type. A partial redesign of the system could decreaseattenuation by an estimated 7dB. In conjunctionwith a new camera objective providing a larger dia-meter collection aperture and infrared-optimizedtransmission, significantly improved SNR valuesat identical dwell times will be achievable. Figure 8shows a summary of these investigations for the twostandard target surfaces by comparing the predictedrange against acquisition time for the current systemand the improved version. These improvements inoptical design provide far greater potential for im-proved range and/or reduced acquisition time forthe scanning sensor than optimization of histogramprocessing.

7. Conclusions

We have demonstrated a scanning time-of-flightdepth sensor with an individual photon-timing detec-tor used in conjunction with galvanometer-mirrorscanning. The sensor has been used in daylight con-ditions to construct depth images of noncooperativetargets at ranges of 325m with μW output power le-vels with centimeter spatial and depth resolution.This low average optical power means that eye-safepower levels can still be maintained despite the pre-dicted increase in laser repetition rate required formore rapid data acquisition, as described in Sec-tion 6. Analyses of the trade-off between targetrange, acquisition time, and optical power have beenmade and future work on the sensor will concentrateon reducing the losses in the receive channel and in-vestigating a larger diameter collection aperture ob-jective. Theoretical modeling of the system hasshown that this approach is capable of the rapid ac-quisition of depth images at kilometer range. Thiswill require further work to increase the maximumusable laser repetition frequency, which is currentlylimited by backreflection from the complex opticalsystem. Work is in progress to operate the sensorat a laser frequency of ∼100MHz. The introductionof a nonperiodic pulse pattern to avoid range ambi-guity between closely spaced optical pulses, as re-cently demonstrated by the group [24], means thatsuch high repetition rates will not compromise therange accuracy or even the maximum depth of fieldpossible in a depth image. This current systemdemonstrates improved depth resolution to otherphoton-counting outdoor depth imaging systemsand comparable, or improved, spatial resolution. De-spite optical losses in the system configuration de-scribed in this paper, this sensor is capable ofoperation at lower optical energies per pixel than si-milar sensors operating using linear multiplication

avalanche photodiodes [11], if operated at the samerange.

The scanning system is inherently flexible and ea-sily reconfigurable, particularly in terms of the objec-tive lens and the detector. The focal length of theobjective lens is a key factor in determining the x −y spatial resolution, and its aperture affects the tar-get return fraction. In terms of single-photon detec-tors, the single-mode optical fiber delivery meansthat other detectors can be easily integrated intothe system. The measurements shown in this paperindicate that potentially improved performance canbe achieved with alternative, low-jitter single-photondetectors such as shallow-junction Si SPADs [30].Other future studies may utilize the low attenuationand eye-safe, atmospheric transmission window at1550nm using new infrared single-photon detectors,such as InGaAs/InP SPAD [13] or NbN supercon-ducting nanowire single-photon detectors [31]. Whilesuch single-photon detector technology may be sig-nificantly less developed than Si, the use of such ascanning platform means that such detectors canbe employed, even when they are necessarily of smallactive area and used in cooled packages. Furtherwork will also investigate the implementation ofdata analysis algorithms suitable for analysis of dis-tributed targets, such as partially hidden targets, asused previously in laboratory-based studies [26].

The work reported in this paper was funded by theElectro-Magnetic Remote Sensing (EMRS) DefenceTechnology Centre, established by the UK Ministryof Defence and run by a consortium of SELEX Sen-sors and Airborne Systems (now SELEX Galileo),Thales Defence, Roke Manor Research and Filtronic.The authors thank their colleagues and collaboratorsfor their valuable contributions and discussions: Pe-ter Heron, Mark Stewart, Ian Chalmers, Fabian Gut-jahr, and Sergio Hernandez-Marin (all Heriot-WattUniversity), Phil Gorman (QinetiQ), Philip Hiskettand Robert Lamb (both SELEX Galileo, Edinburgh),and Michael Wahl (PicoQuant).

References1. F. Chen, G. M. Brown, and M. M. Song, “Overview of three-

dimensional shape measurement using optical methods,”Opt. Eng. 39, 10–22 (2000).

2. M. C. Amann, T. Bosch, M. Lescure, R. Myllyla, and M. Rioux,“Laser ranging: a critical review of usual techniques for dis-tance measurement,” Opt. Eng. 40, 10–19 (2001).

3. C. Mallet and F. Bretar, “Full-waveform topographic lidar:state-of-the-art,” ISPRS J. Photogramm. Remote Sens. 64,1–16 (2009).

4. M. A. Albota, R. M. Heinrichs, D. G. Kocher, D. G. Fouche,B. E. Player, M. E. O’Brien, B. F. Aull, J. J. Zayhowski, J. Moon-ey, B. C. Willard, and R. R. Carlson, “Three-dimensional imag-ing laser radar with a photon-counting avalanche photodiodearray and microchip laser,” Appl. Opt. 41, 7671–7678 (2002).

5. J. J. Degnan, “Photon-counting multikilohertz microlaseraltimeters for airborne and spaceborne topographic measure-ments,” J. Geodyn. 34, 503–549 (2002).

6. C. Ho, K. L. Albright, A. W. Bird, J. Bradley, D. E. Casperson,M. Hindman, W. C. Priedhorsky, W. R. Scarlett, R. C. Smith,J. Theiler, and S. K. Wilson, “Demonstration of literal

6250 APPLIED OPTICS / Vol. 48, No. 32 / 10 November 2009

Page 11: Long-range time-of-flight scanning sensor based on …...Long-range time-of-flight scanning sensor based on high-speed time-correlated single-photon counting Aongus McCarthy,1 Robert

three-dimensional imaging,” Appl. Opt. 38, 1833–1840(1999).

7. R. M. Marino and W. R. Davis, “Jigsaw: a foliage-penetrating3D imaging laser radar system,” Lincoln Lab. J. 15, 23–36(2005).

8. J. S. Massa, A. M. Wallace, G. S. Buller, S. J. Fancey, andA. C. Walker, “Laser depth measurement based on time-correlated single-photon counting,” Opt. Lett. 22, 543–545(1997).

9. W. C. Priedhorsky, R. C. Smith, and C. Ho, “Laser rangingand mapping with a photon-counting detector,” Appl. Opt.35, 441–452 (1996).

10. S. Pellegrini, G. S. Buller, J. M. Smith, A. M. Wallace, andS. Cova, “Laser-based distance measurement using pico-second resolution time-correlated single-photon counting,”Meas. Sci. Technol. 11, 712–716 (2000).

11. B. W. Schilling, D. N. Barr, G. C. Templeton, L. J. Mizerka, andC. W. Trussell, “Multiple-return laser radar for three-dimensional imaging through obscurations,” Appl. Opt. 41,2791–2799 (2002).

12. G. S. Buller, R. D. Harkins, A. McCarthy, P. A. Hiskett,G. R. MacKinnon, G. R. Smith, R. Sung, A. M. Wallace,R. A. Lamb, K. D. Ridley, and J. G. Rarity, “Multiple wave-length time-of-flight sensor based on time-correlatedsingle-photon counting,” Rev. Sci. Instrum. 76, 083112(2005).

13. G. S. Buller, R. E. Warburton, S. Pellegrini, J. S. Ng,J. P. R. David, L. J. J. Tan, A. B. Krysa, and S. Cova, “Single-photon avalanche diode detectors for quantum key distribu-tion,” IET Optoelectron. 1, 249–254 (2007).

14. P. Gatt, S. Johnson, and T. Nichols, “Geiger-mode avalanchephotodiode ladar receiver performance characteristics and de-tection statistics,” Appl. Opt. 48, 3262–3276 (2009).

15. D. G. Fouche, “Detection and false-alarm probabilities for la-ser radars that use Geiger-mode detectors,” Appl. Opt. 42,5388–5398 (2003).

16. J. S. Massa, G. S. Buller, A. C. Walker, S. Cova, M. Umasuthan,and A. M. Wallace, “Time-of-flight optical ranging systembased on time-correlated single-photon counting,” Appl.Opt. 37, 7298–7304 (1998).

17. J. Massa, G. Buller, A. Walker, G. Smith, S. Cova,M. Umasuthan, and A. Wallace, “Optical design and evalua-tion of a three-dimensional imaging and ranging system basedon time-correlated single-photon counting,” Appl. Opt. 41,1063–1070 (2002).

18. C. Niclass, A. Rochas, P. A. Besse, and E. Charbon, “Designand characterization of a CMOS 3-D image sensor based on

single photon avalanche diodes,” IEEE J. Solid-State Circuits40, 1847–1854 (2005).

19. D. T. Neilson, S. M. Prince, D. A. Baillie, and F. A. P. Tooley,“Optical design of a 1024-channel free-space sorting demon-strator,” Appl. Opt. 36, 9243–9252 (1997).

20. C. P. Barrett, P. Blair, G. S. Buller, D. T. Neilson, B. Robertson,E. C. Smith, M. R. Taghizadeh, and A. C. Walker, “Compo-nents for the implementation of free-space optical crossbars,”Appl. Opt. 35, 6934–6944 (1996).

21. R. E. Warburton, A. McCarthy, A. M. Wallace, S. Hernandez-Marin, R. H. Hadfield, S. W. Nam, and G. S. Buller, “Subcen-timeter depth resolution using a single-photon countingtime-of-flight laser ranging system at 1550nm wavelength,”Opt. Lett. 32, 2266–2268 (2007).

22. N. Takeuchi, H. Baba, K. Sakurai, and T. Ueno, “Diode-laserrandom-modulation cw lidar,” Appl. Opt. 25, 63–67 (1986).

23. N. Takeuchi, N. Sugimoto, H. Baba, and K. Sakurai, “Randommodulation cw lidar,” Appl. Opt. 22, 1382–1386 (1983).

24. P. A. Hiskett, C. S. Parry, A. McCarthy, and G. S. Buller, “Aphoton-counting time-of-flight ranging technique developedfor the avoidance of range ambiguity at gigahertz clock rates,”Opt. Express 16, 13685–13698 (2008).

25. A. Berk, L. S. Bernstein, and D. C. Robertson, “MODTRAN: Amoderate resolution model for LOWTRAN 7,” Technical NoteGL-TR-89–0122, available from Geophysics Laboratory/OPE,Air Force Systems Command, Hanscom AFB, Mass., 1989.

26. A. M. Wallace, R. C. W. Sung, G. S. Buller, R. D. Harkins,R. E. Warburton, and R. A. Lamb, “Detecting and character-ising returns in a pulsed ladar system,” IEE Proc. Vision Im-age Signal Process. 153, 160–172 (2006).

27. S. Hernandez-Marin, A. M. Wallace, and G. J. Gibson, “Multi-layered 3D LiDAR image construction using spatial models ina Bayesian framework,” IEEE Trans. Pattern Anal. Mach. In-tell. 30, 1028–1040 (2008).

28. P. J. Green, “Reversible jump Markov chain Monte Carlo com-putation and Bayesian model determination,” Biometrika 82,711–732 (1995).

29. W. P. Cole, M. A. Marciniak, and M. B. Haeri, “Atmospheric-turbulence-effects correction factors for the laser range equa-tion,” Opt. Eng. 47, 126001 (2008).

30. M. Ghioni, A. Gulinatti, I. Rech, F. Zappa, and S. Cova, “Pro-gress in silicon single-photon avalanche diodes,” IEEE J. Sel.Top. Quantum Electron. 13, 852–862 (2007).

31. M. J. Stevens, R. H. Hadfield, R. E. Schwall, S. W. Nam,R. P. Mirin, and J. A. Gupta, “Fast lifetime measurementsof infrared emitters using a low-jitter superconducting single-photon detector,” Appl. Phys. Lett. 89, 031109 (2006).

10 November 2009 / Vol. 48, No. 32 / APPLIED OPTICS 6251