DQNdot: A Deep Learning Framework for Multi-Object Tracking in UAV-Enabled Aerial Surveillance
Neethu Subash, Nithya B S, S Martin Prabhu · 2023
Unmanned aerial vehicle (UAV) object tracking enables various applications in diverse domains, such as surveillance, traffic scrutiny, and search and rescue. However, it presents a formidable challenge that necessitates the proficiency to trail objects under arduous circumstances. In this paper, Deep Q Networks on dynamic object tracking (DQNdot) are presented, encompassing the ensuing components: You only look once (YOLOv8) that spotlights objects in video frames, UAV for Multi-Object Tracking (UAVMOT) traces objects in video frames employing the Kalman filter and Hungarian method. Subsequently, the prioritizing module is also advanced to prioritize things that are more liable to be seamlessly traced. This module harnesses a range of factors to appraise the probability of an object being efficaciously traced, such as its size, speed, and appearance. Deep Q network(DQN) endows the capability to trail resourcefully, learn and adapt to the environment. DQNdot has accomplished 92% accuracy of prophesying the results for pursuit. It transcends all the other frameworks as it can proficiently track as many targets as required.