Smart-Cities: Digitisation of Relief Systems using AI
AI in Disaster Response
Natural Disaster
Natural Disasters are something that are out of reach of our competence. Be it the man-made or natural calamity that occurs, the after effects are quite unmanageable. Current techniques of bringing back life to its normal pace usually consumes quality time and man efforts which eventually paves way for life loss of victims as well as relief groups. The means to automate such after effects have come along from past few decades in which the role of Artificial Intelligence (AI) has been around for a decade now. From means to automate the transport operations to delivery of food items there have been several initialisation taking place. Here I would be taking you through on how existing technologies along with AI techniques could be used for development of a better rescue decision system that could assist the authorities for making quick decisions and direct the relief team for evacuating the victims with less obstructions. Though we have several existing technologies that have been in disaster response operations, here the technique of interest is an Unnamed Aerial Vehicle (UAV).
Digitisation of Relief Systems
Reliable Digital Relief System
The aim of Relief Support Solutions is to digitise the entire critical response routine. A critical element will be the evolution of relief system toward a connected, smart, and highly efficient disaster management ecosystem. The relief systems today are a series of largely discrete, siloed steps taken through manual process at different sectors and finally into the hands of the refugees/victims. Digitisation brings down those walls, and the disaster management becomes a completely integrated ecosystem that is fully transparent to all the players involved in a relief system. With the advent of the digital relief system, silos will dissolve and every link will have full visibility into the needs and challenges of the others involved in it for the same cause. Once built the digital relief system will offer a new degree of resiliency and responsiveness enabling government to provide people with the most efficient relief support solutions.
Here we propose a Reliable Digital Relief System that acts as a virtual support agent for the higher level authorities responsible for decision making during a disaster, critical response team for handling the process as well the victims/refugees to overcome the relief procedures without any time delay. The key elements of this system includes the following,
- Supportive reliable resource to the authorities
- Adaptable to dynamic decisions
- Reduced latency in decision making
- Centralised information sharing centre
- 24/7 available, scalable and omni-channel communication system
Some of the existing Relief systems that are powered by AI and in use real-time are [3],
- Qatar Computing Research Institute (QCRI)
- 1CONCERN
- BlueLine Grid
- AIDR — Artificial Intelligence for (Digital Response)/(Disaster Response)
- StandBy Task Force
AI for Digital Relief System
In case of any disaster, the first step is to formulate a critical response team to help those in distress. Before the team goes into action, it is important to analyse and assess the extent of damage and to ensure that the right aid goes first to those who need it the most.
The analyzation process could be carried out using several techniques.
a) Imaging techniques
- AI techniques such as image recognition and classification can be quite helpful in assessing the damage as they can analyze and observe images from the satellites.
- AI can identify objects and features such as damaged buildings, flooding, blocked roads from these images. They can also identify temporary settlements which may indicate that people are homeless, and so the first care could be directed towards them.
b) Text/Speech processing
- Texts convey the severity of damage caused in a place via social media platforms or through news portals. Voice on the other hand can be received through multiple means such as phone calls, voice messages or video sources.
- Both texts and voice can be another source for analyzation of targeting a particular location based on severity of damage and need for evacuation of people to safer zones.
Relief Support Solutions
a) Digital Report Unit
- Virtual Information Centre: Refugees — QA system to help knowing all obligatory
This system serves as an immediate solution maker to know all basic essentials any affected person would like to know. It may include in getting to know where is the source for getting medications, sanitation care, clothing etc. Such basic necessities would be answered via voice (phone call, google assistance, Siri etc.) or text (web pages or mobile applications). In general, the database will be populated by authorities and also using data from external sources (social media, blogs, news portals, government demographic information, geographic info from external sources etc.) which will be further validated by authorities.
2. Bot for Emergency reporting - Next Generation 911
Under emergency situations when the affected people (refugees) are unable to reach out to any nearby relief team, this voice bot understands the need via call and redirects the message to the nearest relief team for further assistance.
3. Virtual Information Centre: Relief Team - QA system to help knowing essential requirements.
When situations demands taking immediate decisions, this acts as a support agent. This can help the critical response team in getting to know the approximate number of food parcels to be ordered, excessive quantity of medical kits to be shifted, availability of beds in stock and many more. Once again the relief team will be assisted via voice or text. Here information will be provided from government sources as well as statistics made from query made by refugees via virtual information centre and emergency reporting system.
b) Digital Damage Assessment
Below are the sources for assessing damage in a locality
- Through Satellite and Drone Imaging
Satellite images and capturing images using drones are good sources for differentiating areas that are deeply affected from normal ones. Using Deep learning methodologies in vision techniques this could be achieved to target the affected areas immediately for rescue operations to be carried out.
2. Using Internet sources — Social media, news portals, blogs etc.
Rather than relying on various factors Internet alone could be one big reliable solution to target areas that need control over. This can help classifying the areas into sub-areas that needs instant attention. Twitter (collector, tagger) mechanism would be one such solution for understanding the needs in specific localities.
3. Government Database
The mandatory source for damage assessment essentially falls under government database which includes area specific information like population, basic amenities (electricity, drainage facilities, water management etc.) and other economic factors.
The inferred results from these sources would be provided to the authorities (controllers of relief team) by means of a dashboard which includes reports, tables and graphs.
c) Information in Numbers
When information is provided to rescue team in an easily readable format, the time they save can drastically increase the efficiency in delivering goods to the refugees in little time.
- Population at risk with profiling information — Eg: Number of women/babies/children/men
- Determining the need — Eg: Required food quantity, clothing needs etc.
- Formulating plan for critical response team
Let’s now get into how UAVs can solemnly be one good unique technique that could be clubbed with AI for multiple relief support operations.
Unmanned Aerial Vehicles
Yet another source that has been added into the list of providing precise and real-time data are Unmanned Aerial Vehicles (UAVs). While most of us refer to it as drones, these are basically pilotless aircraft that has now been beneficial across multiple domains (Agriculture, Defence, Waste management, Health care, Police department, etc). UAVs have their own pros and cons in its usage as capturing the real time streaming of public has even resulted in security issues. There are separate research and development going on in detecting malicious UAVs and it is expected to reach $1.85 billion by 2024 [1]. While understanding the limitations of UAVs in deployment in real-time there are still certain countries that have even legalized the utilization of UAVs for surveillance. According to a study conducted by the Center for the Study of the Drone at Bard College in 2018 it is believed that two-thirds of public safety departments are liable in incorporating UAVs for carrying out their tasks [2]. In general the global market of UAVs are greatly evolving and is expected to grow dramatically in the next 5 years.
Before getting to know the advantage of clubbing UAVs and AI techniques for disaster response its essential to know the limitations of existing methods that has enforced researchers to incorporate AI for evacuation measures.
Limitations of Traditional methods — Human Intervention
- Minimum accessibility
- Need for skilled man power subject to risks
- Failure of static sources for surveillance (group of people, cameras fixed around the affected zone etc.)
- Lack of in depth analysis due to unavailability of higher-detail data (images, video frames etc.)
- Absence of quick reflex actions on event occurrence
Selection of suitable UAVs
For real-time deployment of the suggested approach it is important to choose the right UAV suitable for disaster response. UAVs come under different size, weight, operating altitude, range and payload in general. After a good research, I came to know that Mini and Micro UAVs are the ones that have been currently used for disaster management [1]. While Mini UAVs or drones are capable of small payloads and altitude coverage upto seven kilometres, micro UAVs fall under the same altitude range but with limited payloads as their main purpose would be surveillance. Wi-Fi First-Person View (FPV) channel with Return-To-Home (RTH) or Follow Me facility would help in automatic manoeuvring of the drone to a predefined home target and landing based on a GPS device. While there is a loss of connection between the Ground station and the UAV, RTH will automatically return back which simply signifies the reason behind its name. The Follow Me is basically controlled by the drone operator and essentially uses Computer Vision technologies for object detection and segmentation [1].
Advantage of UAVs in Disaster Relief
- Avoids exposing humans to unnecessary and unpredictable hazards
- Being dynamic and flexible in nature allows UAVs to capture the event in multiple angles better than static surveillance cameras and manned aircraft
- The size of the UAVs provides the advantage to access the damaged infrastructure better than the manned aircraft
- Provide access to the CBRNE events without compromising the safety
- Provides high scalability, reliability and simplicity in deployment
Security Conditions to be Addressed
- Sending data in real-time without storage to avoid the ceasing of UAVs by anomalies
- Building hyper secured wireless network between UAV and base station
- Role based access to UAV
- Authentication protocol
- Encrypted up-link and down-link data
After the selection of suitable UAVs for the evacuation operation it is essential to make use of historical data along with the current event that the UAVs capture. Mix up of real-time and past data can help authorities to make quick decisions and guide the relief team accordingly. This could save time and hence save lives of both the rescue team as well as refuges to find a way out during the outbreak of any calamity for that matter. A sample outlook to the layout that could possibly be used during rescue operations have been illustrated below. This is one of Arnekt’s reputed ideologies in using AI for relief operations. Having its own restrictions to exclude the techniques used behind I will not be taking you all through an in-depth explanation of the architecture. In short Deep Learning algorithms have played a major role in projecting the damaged areas from the unaffected ones. This prototype has its strength of automating the entire relief operations by providing quick options to carry out the evacuation measures.
In short the overall flow goes like this. During a crisis, the data from the UAVs that are controlled using mobile phones or a computer from the ground will be sent to the Secured Evacuation Platform hosted in Cloud. This simply means that the proposed AI routine has been trained on the actual data (unaffected/affected) of the particular locality that has been affected and hosted in the cloud. During run-time data from UAVs and sensors are sent to the module in cloud that generates importance score based on the priority of the affected area along with ways to find the route to safer and faster evacuation procedure. The results are obtained to the ground station via dashboards and graphs. These results help in making quick eradication measures and the same will then be sent to the authorities in charge of taking decisions. Once authorities approve the plan, the recovery team is all set to start their evacuation mechanism.
Arnekt has not only come up with solutions to relief operations but also come up with a justification as to why such outputs occur during the selection of particular model. More on it can be found on my next blog on Transparency in AI — SpyFlow. The same technique has been applied to the current relief support architecture selection. The above illustrated prototypes are exclusively designed in Arnekt in pure interest towards applying AI in bridging the time gap between relief operations and digital technologies. The suggested design flow proves to be a reliable if used in the right way (choosing the suitable algorithm).
