AI and Drones for Search and Rescue

How INGD and WFP combine drone imagery and rapid AI for large-scale Search-and-Rescue operations


Tropical Cyclones  Idai  and  Kenneth  hit Madagascar, Mozambique, Malawi and Zimbabwe in March and April 2019. Together they dumped several months’ worth of water on the area, causing devastating floods in addition to wind damage.

A race against time was on to search for and then rescue thousands of people dispersed over huge areas, with little food, water or shelter.

The  United Nations World Food Programme  has been operating across Southern Africa for over two decades supporting food delivery to some of its most vulnerable populations and regularly puts its formidable logistical capacity and aerial assets to use in search and rescue (SAR) operations.

Given the wide areas involved however its helicopters end up flying many expensive hours before anyone is sighted.

In 2021 the  WFP Drones  team based in South Africa teamed up with Mozambique's National Disaster Management Institue (INGD) to study the potential for cheaper Unmanned Aerial Vehicles to quickly carry out the search element.

The goal is to test the feasibility of replacing the use of helicopters in the search phase of SAR, using drones instead to identify the location of people needing rescue, and only then sending the helicopters and boats directly to them. Through this effort, WFP and INGD aim to use their rescue resources more efficiently and thereby save more lives.


Synthetaic's Rapid Automatic Image Classificaiton (RAIC)

Start by showing RAIC an example of what you're looking for then launch a quick search for similar features across the imagery set.

A deep search will produce a context map that clusters image tyles by level of confidence in similarity to the original search item.

The centroids of the image tiles can be exported to an ArcGIS Feature Layer for further mapping. The Feature Layer contains an attribute for the total number of people detected in the tile.

DRONE TASK MANAGEMENT WORKFLOW

The image analysis described above assumes of course that you already have access to drone imagery. But managing the image acquisition process upstream of this requires robust planning and monitoring tools, especially in complex emergencies where multimple actors and multiple Areas of Interest are involved.

We use a combination of Survey123, ArcGIS FieldMaps, ArcGIS Dashboards, SiteScan and ArcGIS Web AppBuilder to monitor Areas of Interests (AOIs) through their life cycle.

The Emergency Coordinator starts by creating and assigning an AOI. This includes information on the specifications of the imagery required.

Tasks, their assigness and their status are made visible to the Emergency Coordinator on a dashboard.

Pilots inspect their assigned tasks on ArcGIS FieldMaps. Once onsite they mark the task as 'In Progress'.

The polygon can be ingested directly in SiteScan for flight planning.

Once the flight is complete the pilot can notify the Emergency Coordinator by setting the status to 'Imagery acquired'.

Once the Emergency Coordinator has validated the final product the AOI can be marked as 'Complete'.