Occluded Target Surveillance A Topological Perspective on Intermittent Target Tracking with Lyapu...

This work introduces a novel framework for image-based tracking systems, addressing scenarios where the tracking agent needs to discontinue tracking the target either due to the need to fulfill other tasks or the target becoming obscured. The proposed approach deploys a Lyapunov-based Deep Neural Network (Lb-DNN) to learn the dynamics of the target when visible, and to predict its future trajectory when not visible. To ensure that target tracking resumes, a topologically inspired method is proposed, using the predicted trajectory of the target. This method informs the tracker agent about the duration it can suspend tracking the target and specifies a pose for the camera for guaranteeing that tracking resumes at some later time. Simulation results are provided to demonstrate the efficacy of the developed methodology for the task of tracking a target intermittently, where the topologically-inspired camera placement algorithm is successfully deployed to reacquire tracking of the target.