PERECO - Personalized Emergency ResponseEcosystem

Chiraag SumanthPES Institute of Technologychiraagsumanth@gmail.com

Aditya DalwaniPES Institute of Technology

adit.39@gmail.com

Arvind SrikantanPES Institute of Technologyarvind.s01994@gmail.com

Satyanarayana SrinivasPES Institute of Technology

satyasrinivas@pes.edu

Abstract—An important drawback of emergency managementplans is that their major focus is on the relief, rescue andrehabilitation of victims after the disaster has struck. This hasinherent complexities as any post-disaster region or communitywould suffer from infrastructure problems such as incapacitatedcommunication channels, large scale electricity outages, shortageof essential supplies such as food and water, damaged andblocked roads, bridges, and highways. Such a situation, commonin most post-disaster regions paralyse the entire region andoverwhelm the relief and rescue authorities hampering theireffective operation. Therefore such relief, response and rescuemethods that come into action during the aftermath of disasterstill witnesses considerable loss of human-life and property. Inthis work, we define a new disaster response mechanism, that ismobile-deployed and cloud-driven, and works on the principleof prevention is better than cure, attempting to overcome thesedrawbacks of current emergency management plans.

Keywords—emergency management, rescue, response, relief,prevention, mobile-deployed, cloud-driven

I. INTRODUCTION

Emergency management (or Disaster management) [2] in-volves the creation of plans through which communities reducevulnerability to hazards and cope with disasters. Disastermanagement does not avert or eliminate the natural disastersor threats, instead it focuses on creating plans to decreasethe impact of natural disasters. Failure to create a plan couldlead to damage to assets, human mortality, and lost revenue.The Federal Emergency Management Agency (FEMA) has setout a basic four-stage vision of emergency management andplanning flowing from mitigation to preparedness to responseto recovery:

• Mitigation- Personal mitigation is a key to nationalpreparedness. Individuals and families train to avoidunnecessary risks. This includes an assessment ofpossible risks to personal/family health and to personalproperty, and steps taken to minimize the effects ofa disaster, or take procure insurance to protect themagainst effects of a disaster. Without mitigation actionsagainst future emergencies or disasters, we jeopardizeour safety, financial security and self-reliance. Anexample could be constructing of earthquake-resilienthomes in the aftermath of a major earthquake ina particular region, based on the premise that thiswould reduce damage to life and property if a futureearthquake were to occur in that region.

• Preparedness- Preparedness focuses on preparingequipment and procedures for use when a disaster

occurs. This equipment and these procedures can beused to reduce vulnerability to disaster, to mitigate theimpacts of a disaster or to respond more efficiently inan emergency. The basic theme behind preparedness isto be ready for an emergency and there are a numberof different variations of being ready based on anassessment of what sort of threats exist. Preparednessmeasures can take many forms ranging from focusingon individual people, locations or incidents to broader,government-based ”all-hazard” planning.

• Rescue- The response phase of an emergency maycommence with Search and Rescue but in all casesthe focus will quickly turn to fulfilling the basichumanitarian needs of the affected population. This as-sistance may be provided by national or internationalagencies and organizations. Effective coordination ofdisaster assistance is often crucial, particularly whenmany organizations respond and local first-respondersand rescue agencies have been overwhelmed by thedemand or diminished by the disaster itself. Addi-tionally, because this phase is practised soon afterthe disaster has struck, authorities are faced withextremely difficult challenges as most of the supply,communication and transportation infrastructure of theaffected region will be detrimentally incapacitated andunfit for regular use.

• Recovery- The recovery phase starts after the immedi-ate threat to human life has subsided. The immediategoal of the recovery phase is to bring the affected areaback to normalcy as quickly as possible and mitigatethe outbreak of any major post-disaster hazards suchas disease epidemics. This phase is also faced withsimilar challenges as the rescue face primarily due tothe incapacitated transport, communication and otherlocal civic infrastructure and resources.

It is important to note that one of the most importantaspects of emergency or disaster management is Prevention-This phase was not in the original four phases of emergencymanagement as described above. However, it is a crucial partof any disaster management plan and focuses on preventingthe human hazard, primarily from potential natural disasters orterrorist attacks. Preventive measures may be taken on both thelocal, national and international levels. In January 2005, 168Governments adopted a 10-year global plan for natural disasterrisk reduction called the Hyogo Framework. Unlike the otherphases, the prevention phase is practised before the disaster hasactually occurred, and is thereby provided the opportunity for

our solution, which focuses on natural disaster management, toleverage all communication, transport, supply and other civicinfrastructure and resources that is still intact in that regionbefore the disaster actually occurs and consequently attemptsto reduce casualties of both human-life and material-assets.

A. Case Study: Uttarakhand Floods of 2013, India

At the peak of the monsoon season the north Indian stateof Uttarakhand came face to face with with what was tobecome India’s worst natural diaster since 2004.The 2013North India floods” as it is termed, left tens and thousandsof inhabitants as well as pilgrims stranded or swept away dueto the floods, in addition to inflicting considerable damageto assets, property and businesses. The main day of flood issaid to be 16 June, 2013. As of 16 July 2013, according tofigures provided by the Uttarakhand government, more than5,700 people were ”presumed dead.” There are several reasonsthat can be attributed to this disaster. However, one of themost glaring irregularity is that the Indian MeteorologicalDepartment had issued advance warnings as early as 13 June,2013 that flash floods are imminent and had even identifieddistricts and regions as very high-risk zones. These warningswere not disseminated to local inhabitants, pilgrims or touristsin that region and neither did any local or state governmentagency take action based on these warnings. This demonstrateda complete lack of co-ordination between various governmen-tal agencies, and consequently, no actionable information orwarnings was made available to the thousands of innocentvictims of this unfortunate incident, who were completelyunprepared to face such a devastating disaster.

During the aftermath of the disaster, the entire region wascut-off from the rest of the country as all road-transport,communication and supply lines were destroyed. This not onlydelayed relief, rescue and rehabilitation operations, but alsomade it extremely challenging and difficult as most areas wereonly accessible by air. Thus, such a reactive response in theaftermath of the disaster could not prevent the extremely largelosses of life and property that were caused as a direct result ofthe disaster, though rescue efforts were successful in rescuinglarge numbers of stranded pilgrims, tourists and locals, whohad survived the flash floods.

Thus, a preventive or proactive solution that could lever-age the early warning data made available, process that anddisseminate actionable information in mass, to the people onthe ground could have possibly led to thousands of more livesbeing saved and also reduce property and asset damage causedas a result of this natural disaster.

II. SYSTEM DESIGN

In this section, we will present the overall design ofour mobile-deployed, cloud-driven solution called PERECO.Figure 1 shows the overview of the proposed system.

The system components are described below.

1) Our system is designed to ingest heterogeneous tex-tual weather data from various sources such as me-teorological departments and private weather-forecaststations. Any warning issued will be interpreted byour cloud backend and trigger an emergency action

Fig. 1. PERECO System Overview

sequence which is described as follows. Our systemis designed to analyse any textual data format suchas JSON, XML, Plain text, PDF or HTML.

2) One of the most crucial parts of our entire system isthe cloud backend. This component is responsible fortying our end users and data sources. It also performsseveral critical functions.

• Geo-fencing: A geo-fence [1] is a virtualboundary defined by a series of latitude andlongitude co-ordinates around a particular re-gion. In the weather or advance warning data,suburb or district names are identified aspotential high-risk zones. Based on this ge-ographical data the system receives, the geo-fencing component outputs a series of latitudeand longitude co-ordinates that together geo-fences a potential danger or high-risk zone.Additionally, with the data received, this com-ponent also identifies the nearest ”safe-zones”outside of the geo-fenced danger zone, whichis considered as a safe region for our users.

• Routing: This is the component that indi-vidually determines the most optimal escaperoute for each user in the danger-zone, therebydirecting them to safety. The algorithm is

described in detail in the following ”Imple-mentation” section.

• Real-time tracking and monitoring: Thecloud backend is constantly updated with lo-cations of our users, from the mobile applica-tion running on the end-user phones. Thus thebackend always has a big-picture of the entireon-ground situation at any instant of time.The mobile application also has an option toreport obstacles. On receiving such a triggerfrom any user mobile application, the cloudbackend engages the re-routing component,which is described in the following ”Imple-mentation” section.

• Accident detection and reporting: A uniquefeature of our end-user mobile application isthe ability to detect any road-accidents theuser gets into, irrespective of the vehicle typeand any other external sensor requirement.Also, if another user spots an accident, theycan report it via the mobile application toour backend. On receiving such a triggerfrom any user mobile application, the cloudbackend engages the automated accident re-sponse component, which is described in thefollowing ”Implementation” section

3) End-user mobile application: This is the componentof the system that runs on the mobile devices [3] ofthe end-user. We collect basic user information andalso the contact details of an emergency contact orloved one from the user, the first time they use the ap-plication. The application itself is designed to be non-intrusive and run in the background, and designed fol-lowing strict developer guidelines to minimize excessbattery drainage issues. The mobile application willautomatically launch (no end-user intervention re-quired), with an auditory stimulus to catch the user’sattention, whenever it receives an emergency actionmessage from the server. This communication canhappen via the Hyper Text Transfer Protocol (HTTP)or an SMS message. The user is then required tofollow on-screen instructions to proceed immediatelyto safety. In transit, the user can manually report anyobstacles, roadblocks or accidents, in which case theyare quickly re-directed to an alternative safe-zone. Ifthe user is involved in any road accident, the mobileapplication has an automated accident detection andresponse component and the backend is immediatelyand automatically notified of the accident. Emergencyresponse authorities can then be notified without anyfurther delay to proceed to rescuing the accidentvictim.

III. IMPLEMENTATION

We now present the key algorithms that we have developedas part of PERECO.

A. Routing Algorithm

1) Based on location [4] reported from user’s device,those in the geo-fenced danger zone are immediately

identified. Also, safe zones are identified around thegeo-fenced danger zones using the same data.

2) These users are then clustered into groups. The num-ber of clusters in a subregion within the danger zoneis directly proportional to current user count in thatsubregion. The clusters geographic radius is restrictedto 10 kms. This upper limit is imposed to ensure thealgorithm does not irrationally increase the size ofthe cluster.

3) Every cluster thus obtained is a part of a graph Gwith the cluster itself as the source vertex and thesafe zones as destination vertices. The various pathsare edges of this graph with weights of the edgesbeing proportional to the distance.

4) The number of such graphs will be equivalent to thenumber of such clusters, each cluster being the sourcevertex of its respective graph.

5) In each graph G do the following:• The shortest path is found by selecting the

edge with the least weight.• When the capacity of safe zones is completely

exhausted remove the vertex representing thissafe zone from all the graphs.

• Therefore, each user in every cluster is nowdesignated a safe zone and an optimal routeto that safe zone.

6) During transit, in case any user reports a path-blockage:

• Remove the edge corresponding to the pathreported blocked from all graphs.

• Reassess an alternative optimal safe zone foreach user on that path using the steps dis-cussed above. Notify user of new directionsto safe zone.

• Notify all users, in real-time via the mobileapplication, who are currently on that pathwith directions to the alternative safe zone.

B. Accident detection and response

Road accident fatalities are another major cause of humanlives being lost. According to the World Health Organization,approximately 1.24 million deaths occurred on the world’sroads in 2010. The reasons for such high road-accident mor-tality rates are as follows.

• Delay in reporting of the accident to emergency re-sponse authorities.

• Ambiguity in the exact location of the accident.

• Delay in locating and intimating the closest medicalcare facility.

Our aim with this component of PERECO is to eliminatetime delays and human intervention involved during key stagesof an accident emergency response. Thereby, the developedsystem attempts to completely automate the entire process ofdetection, reporting and initiation of rescue soon after a roadaccident has occurred, striving to make medical care availableto the victim(s) within the crucial first hour after the accident,termed as the ”Golden Hour”.

Fig. 2. PERECO: Accident detection and response

Additionally, users who witness road-accidents at theirlocation can also choose to manually report the same via themobile application, which would initiate a similar response,if not already triggered automatically by the accident victim’smobile device itself.

The premise on which this component is built on is that,in the event of an accident, the occupant’s mobile device willexperience the same forces as that of the vehicle’s occupant.The accident detection algorithm, again working completelyin the background in a non-intrusive way, uses the mobiledevice’s on-board accelerometer sensor [6] to detect suddenacceleration changes, and if available, uses the on-boardgyroscope [5] to detect sudden changes in orientation. Thesystem is designed to initiate an automated accident responsemessage if the change in acceleration is upward of a G-forceof approximately 60Gs , which is the G-force required todeploy a vehicle’s airbags. A buffer of 5Gs introduced toaccommodate different accelerometer accuracies .Thus, thisacceleration filter, coupled with a simultaneous and suddenorientation change as measured by the gyroscope, if availablewill ensure no false positive alerts are raised, such as duringsudden braking or dropping of the mobile device the head-height of the user. These are only known to produce G-forcesof not more than 4Gs in previous experiments [9], and hencewill not trigger the emergency accident response message fromthe mobile application. Figure 2 depicts the working of thiscomponent.

The cloud backend receives regular updates about thelocation of emergency response vehicles through their GPSunits or a mobile application that can be carried by theoperator. The list of nearby hospitals is available through theGoogle Places API. Thus, on receiving an accident emergencytrigger, the nearest emergency response unit is immediatelynotified of both the accident location and is given details of thenearest hospital to where the victim is to be transported. Theemergency contact or loved one of the victim are also notifiedwith these details. The hospital may also be notified of theincoming victim. therefore this component achieves completeautomation in key stages of accident detection, reporting, andresponse and ensures that medical care reaches the accidentvictim as soon as possible.

IV. SIMULATION AND RESULTS

This section presents a simulated run of the algorithm usingweather warning data for the Uttarakhand floods of 2013 inIndia, the facts of which were already presented in the sectionabove.

Figure 3 shows the early warning text data (PDF format)retrieved from the Indian Meteorological Department on 13June 2013. The floods occurred on 16 June 2013.

Fig. 3. Advance weather warning. The messages marked red imply immediateaction necessary. The yellow and green areas are considered safe for humanhabitation.

Based on the regions considered to be in danger, the geo-fencing component created geo-fences and determined thedanger-zones and surrounding safe-zones after analysing andparsing the data presented. Figure 4 shows this visualizationon a map of the region.

Fig. 4. The danger zones, marked in red are geo-fenced. There are two safe-zones, marked in green, are the major centres in the neighbouring districtswith the capability and resources to handle the incoming evacuees.

The clustering is then performed to calculate the most

optimal safe-zone and route for each user in the geo-fenceddanger zone.

Fig. 5. The clustering is performed as per the algorithm. each user is assignedone of the safe zones and directed to proceed there via an optimal route.

The mobile application running on the user’s mobile deviceis launched automatically on receiving an emergency actionmessage from the cloud back-end. the user is then directed tothe nearest safezone as shown in Figure 6.

Fig. 6. The mobile application is automatically launched with an auditoryand visual warning. The user is then directed to take the optimal route to theassigned safe zone.

Thus, our system works according to the algorithm de-signed and successfully delivers timely, actionable informationto the mobile devices of the users in the danger zone, in order

for them to reach safety well before they find themselves inan emergency situation.

V. CONCLUSION AND FUTURE WORK

With the advent of mobile smart-phone technology, it hasbecome increasingly easy for the dissemination of data tomobile smart phones, and processing of that data to giveactionable information to the users. This work attempts todefine a new disaster response mechanism that works onthe principle of prevention is better than cure and focusesmainly on how to safeguard human life by leveraging allcommunication and transport and other civic infrastructure andresources that is still intact in that region, well before thenatural disaster occurs by utilizing information advanced earlywarning systems already in place at meteorological stationsworldwide and a cloud-powered smart phone mobile appli-cation that allows for actionable information to be deliveredto users in a timely manner, through which several livescan be saved. Additionally, this work proposes a method,using the same mobile application and cloud-powered back-end to completely automate the process of detection, reportingand response to road accidents, considerably reducing delaysin medical care reaching victims, thereby maximizing theirchances of survival. This overall framework that is developedis termed as PERECO - Personalized Emergency responseEcosystem.

Our future work includes research on developing a rec-ommender system by exploring the utilization of the geo-location data [7] we collect, to help automatically identifyroad-accident hotspots and evacuation route bottlenecks [8].Additionally the analysis of this data facilitates the optimiza-tion of the placement of emergency response units and medicalcare facilities to minimize the time involved in deliveringemergency medical care to any victim in that particular region.Since the granularity of our data is considerably localized,region-specific recommendations can be made, depending onthe local conditions and resources available. Therefore, overtime, with this actionable information at our disposal, ourroads can be made safer for both pedestrians and motoristsand our cities and towns could be much better prepared andequipped to handle any large-scale evacuation scenario, beforean impending disaster situation.

REFERENCES

[1] D. R. Sanqunetti, “Implementing geo-fencing on mobile devices,” 2004,uS Patent 6,721,652.

[2] A. Dyregrov, “Caring for helpers in disaster situations: Psychologicaldebriefing,” Disaster management, vol. 2, no. 1, pp. 25–30, 1989.

[3] S. Chapman and W. Schofield, “Lifesavers and samaritans: emergencyuse of cellular (mobile) phones in australia,” Accident Analysis andPrevention, vol. 30, no. 6, pp. 815–819, 1998.

[4] J. P. Munson and V. K. Gupta, “Location-based notification as a general-purpose service,” in Proceedings of the 2nd international workshop onMobile commerce. ACM, 2002, pp. 40–44.

[5] C. Barthold, K. P. Subbu, and R. Dantu, “Evaluation of gyroscope-embedded mobile phones,” in Systems, Man, and Cybernetics (SMC),2011 IEEE International Conference on. IEEE, 2011, pp. 1632–1638.

[6] L. Sun, D. Zhang, B. Li, B. Guo, and S. Li, “Activity recognition onan accelerometer embedded mobile phone with varying positions andorientations,” in Ubiquitous intelligence and computing. Springer, 2010,pp. 548–562.

[7] Y. Zheng, Q. Li, Y. Chen, X. Xie, and W.-Y. Ma, “Understanding mobilitybased on gps data,” in Proceedings of the 10th international conferenceon Ubiquitous computing. ACM, 2008, pp. 312–321.

[8] M. Dilley, Natural disaster hotspots: a global risk analysis. World BankPublications, 2005, vol. 5.

[9] J. Barnes, A. Morris, B. Fildes, and S. Newstead, “Airbag effectivenessin real world crashes,” 2001.