Closed Loop Performance Enhancement in Real-Time Object Tracking with Multi Tracker Fusion

Closed-loop object tracking is a broad area of research with applications in surveillance, space, defense, and other fields. A tracking system consists of a multi-axis gimbal system with mounted imaging sensors in the UV, visible, and infrared spectra. The objective of the system is to track an object or area of interest using a closed-loop servo system. Object tracking algorithms have seen a significant improvement in performance with the rise of AI and deep learning implementations. However, the main challenge faced by most systems in real-time implementations of these algorithms is the limited computational power and stable closed-loop tracking. These algorithms require high-performance Graphics Processing Units (GPUs) and processors for real-time applications, which is sometimes not feasible in existing systems. Additionally, for stable closed-loop tracking, the error generated by these algorithms must be regulated for the system to effectively follow the object of interest. We propose a new method for the real-time parallel implementation of tracking algorithms to achieve accurate and stable closed-loop tracking. The method combines two algorithms to ensure stable real-time closed-loop performance. The results demonstrate a significant improvement in both closed-loop tracking and object tracking performance.