Real-time Multi-Object Tracking for Rescue Operations
SORT AND DEEP SORT: TO SUPPORT RESCUE EFFORTS BY TRACKING PEOPLE AND THEIR MOVEMENTS IN DISASTERS
In response to the devastation caused by Tropical Storm Belal in Mauritius, I aim to leverage my computer vision expertise to aid rescue operations. Observing the urgent need to locate individuals trapped by the storm, particularly those stranded in vehicles amidst floodwaters, I propose developing a system that employs surveillance and drone technology. This system would monitor disaster-stricken areas in real-time, accurately tracking people in need of rescue to enhance emergency response efforts.
The project focuses on creating tracking algorithms that can effectively follow the movement of individuals during disasters. Initially, I explore the SORT algorithm, which utilizes the Hungarian algorithm for the association of tracks and the Kalman Filter for motion modeling. However, SORT struggles with occlusions and frequent identity switches. To address these issues, I progress to implementing the Deep SORT algorithm. Deep SORT not only models motion but also incorporates appearance information through a Siamese network, significantly improving tracking accuracy during complex scenarios.
The objective of this initiative is to improve the capabilities of disaster response teams. By providing a reliable method to monitor and track individuals affected by disasters like cyclones or floods, the project aims to enhance situational awareness, coordination among responders, and ultimately save lives by ensuring timely and efficient rescue operations.
