Multi-object Segmentation and Tracking System on Zynq UltraScale+ MPSoC

Multi-object segmentation and tracking is one of the prominent applications and a challenge for edge computing systems. This paper addresses the critical challenge of multi-object segmentation and tracking, leveraging the capabilities of the Zynq UltraScale+ MPSoC ZCU104. The proposed solution incorporates state-of-the-art object detection through YOLOv8, a highly efficient and accurate deep learning model. The fusion of YOLOv8 with different tracking algorithms such as Deep OC-SORT and StrongSORT enables robust tracking of multiple objects over time tackling problems such as occlusions and re-identification. The integration of complex deep learning models such as YOLOv8 and heavy-weight tracking algorithms with the Zynq UltraScale+ MPSoC ZCU104 platform aims to explore the capabilities of the ZCU104 board and achieve a resource-efficient multi-object segmentation and tracking. The use of the Zynq UltraScale+ MPSoC ZCU104 board aims for the prominent application of the board in ADAS features such as object detection, segmentation, and tracking. So the paper focuses on the use of edge computing devices for the application of object segmentation and tracking. The outcomes of this system hold significant implications for applications such as video surveillance and autonomous systems.