FURI | Spring 2022

Object Detection and Tracking for a Pursuit-Evasion Scenarios

Counter-UAV systems are growing popular day by day. As the rise of drones has its benefits, but also raises the risks of potential threats when handled in unlawful activities or involved in causing harm to the citizens. Real-time, recurrent, regression-based tracking, a fast yet accurate network for generic object tracking, was developed for drones to track objects in real time. The custom tracking algorithm is trained to detect more than 40 different objects and can detect multiple objects in the same frame. Customizing the YOLOv3 algorithm with better performance extended with instance segmentation removes two of its weaknesses: a large amount of rewritten labels and inefficient distribution of anchors. The bounding polygon is detected within a polar grid with relative coordinates that allow the network to learn general, size-independent shapes. The network produces a dynamic number of vertices per bounding polygon.