Real-Time Multi-object Detection and Tracking on Zynq UltraScale+ MPSoc

The increasing demand for real-time and accurate multi-object tracking (MOT) systems has prompted advancements in computer vision and hardware acceleration technologies. This paper introduces a holistic approach to multi-object tracking implemented on a Zynq UltraScale+ MPSoc ZCU104 Evaluation Kit. The proposed system harnesses the YOLOv8 object detection model from Ultralytics for robust and efficient object detection. Additionally, it integrates the OC-SORT and ByteTrack algorithms to ensure reliable multi-object tracking. The choice of the YOLOv8 model stems from its cutting-edge performance in object detection, offering a crucial balance between high accuracy and speed essential for real-time applications. The OC-SORT algorithm, an enhancement of SORT, is a real-time, online, and user-friendly multi-object tracker based on a pure motion model. ByteTrack, recognized for its precision, speed, and efficiency, serves as a real-time, high-performance tracker adept at tracking small, closely spaced objects. Leveraging the ZCU104 Evaluation Kit as the hardware platform, experimental results affirm the effectiveness of the proposed system in achieving accurate multi-object tracking, rendering it suitable for diverse applications such as surveillance, autonomous vehicles, and robotics.