PRELIMINARY DESIGN REVIEW HiMax: Facial Biometrics With a CogniMem Device (Neural Processor)
-
date post
22-Dec-2015 -
Category
Documents
-
view
216 -
download
0
Transcript of PRELIMINARY DESIGN REVIEW HiMax: Facial Biometrics With a CogniMem Device (Neural Processor)
PRELIMINARY DESIGN REVIEWHiMax: Facial BiometricsWith a CogniMem Device
(Neural Processor)
Presentation Overview:
Team and Member Introduction Brief Project Overview Approaches
– Phase 1– Phase 2– Phase 3
Possible Problems (guarantee there will be some!) End of Semester Project Goals Project Timeline Presentation Summary References Q&A session
Advisor: Dr. Tep DobrySub Advisor: Dr. Neil ScottDr. Winyu Chinthammit
Members: Raymundo Flores James Cuaresma
PRELIMINARY DESIGN REVIEWHiMax: Biometrics
With a CogniMem Device(Neural Processor)
Project Overview:
This technology is fairly new, so we propose:
1. Research methodology for an “efficient training” of the CogniMem Neural processor.
2. Concurrently, develop “reproducible procedures” for a "high-level confidence" for “Facial Biometric recognition” application in a static physical environment.
3. As needed, create hardware interface with the device; develop software for a particular application.
Block Diagram:Controlled Environment Diagram
Side View
Top View
Standard Settings: - Method of training: Easy-Video Recognition - Image Recognition Type: Moderate - Region Of Interest (ROI): (244,126)-(208,195) - Light elevations: 65.5”
Variable Settings - Refer to Scripted Matrix Call-out
Block Diagram:Controlled Environment – Scripted Matrix Call-out
TEST 1: Lighting Variation Effect
Block Diagram:Controlled Environment – Scripted Matrix Call-out
TEST 1: Lighting Variation Effect (Continuation)
Block Diagram:Controlled Environment – Scripted Matrix Call-out
TEST 2: External Facial Accessories Variation Effect
Block Diagram:Controlled Environment – Scripted Matrix Call-out
TEST 3: Facial Change and Facial Emotions Variation Effect
Block Diagram:CogniMem Device and Applications
Block Diagram:CogniMem Device (Details)
Block Diagram:Laptop Temporary Setup & Training (CogniSight Control Panel)
Approaches:
Phase 1 (Completed) Create a “controlled environment” to characterize the Biometric limits of the
CogniMem device with respect to known environmental changes (more elaboration is just a sec).
“Lesson learned “ comprehensive review of result (in-progress).
Phase 2 With lesson learned in Phase 1: Will implement a controlled-group “open
environment” characterization/testing of the device (more elaboration is just a sec).
Phase 3 Verification and “variable-environment” testing of the controlled-group (8
individuals) to the general population False-Positive
Control group (thru external facial manipulation) to deceive & create “Positive Unknown” identification
Phase 1: Results
Result Discussion:o Lighting Variation Effectso External Facial Accessories Variations
Effecto Facial Changes Effectso Facial Expression Effects
Phase 1: Lesson Learned
o Perform training with “sufficient lighting”o “Zoom-in” ROI size to facial contour.o If possible, set camera configurations to
“manual mode” (i.e. shutter speed, gain, etc.)o First neuron committed should be on the
background environment.o CogniMem device is somewhat unstable.
- Froze 3X (after 10, 2 & 40 minutes of operation).
- Uncontrolled neuron assignment.
What’s next?Phase 2 & Phase 3
Phase 2With lesson learned in Phase 1: Will
implement a controlled-group “open environment” characterization/testing of the device (more elaboration is just a sec).
Concurrently, research cause for theo CogniMem to “freeze”o Uncontrolled neuron assignmento USB port assignment erratic
What’s next?Phase 2 & Phase 3
Phase 3Verification and “variable-environment”
testing of the controlled-group (8 individuals) to the campus general population False-Positive
Control group (thru external & facial manipulation) to deceive & create “Positive Unknown” identification
Problems (Yes, we have them):
Hardware & Softwareo CogniMem to become unresponsiveo Uncontrolled neuron assignments
(without any keyboard or mouse activity); normally happens right after the CogniMem device becomes unresponsive.
o USB port assignment erratic
End of Semester Project Goals:
Our projected goals are to: Develop efficient training methodology of the CogniMem
device. Develop “reproducible procedures” for a "high-level
confidence" Biometric recognition training. Have a working model for a static Facial Biometric
identification that is reliable. Hopefully, find solutions to the hardware and software
problems we are experiencing to have a more stable system.
EE496: Lay the ground-work for a “dynamic” Facial/Body Biometric follow-on project.
Project Timeline
9/15-9/26 9/27-10/10 10/11-10/24 10/25-11/7 11/8-11/21 11/22-12/5 12/6-12/12
Presentations
Proposal Complete
PDR Complete Complete
CDR
Final
Tasks
Obtain Lab Complete
Set up Lab Complete
Familiarize with Complete Complete
Cognimem (James) Complete Complete
Paliminary testing Complete Complete
Training on volunteers Started
Known to unknown ratio testing
?Any Questions?