Multi-Target Instance Segmentation and Tracking Using YOLOV8 and BoT-SORT for Video SAR

Abstract

Moving target detection and tracking is a crucial application field of video SAR. It can offer significant support for military intelligence reconnaissance and civil monitoring. Compared with the traditional moving target detection and tracking methods, the deep learning-based moving target detection and tracking methods can more fully extract the moving targets' shadow features in video SAR frame images to improve detection and tracking accuracy. This paper proposes a multi-target instance segmentation and tracking framework based on the YOLOv8 model and the BoT-SORT algorithm for video SAR. Firstly, the detection effects of YOLOv8-seg models with different sizes are analyzed, and the most suitable detection model for video SAR moving targets is selected. Secondly, the BoT-SORT algorithm is used to track the targets detected by the detector to complete the whole detection and tracking process. Based on the experimental results, it has been determined that the proposed framework is highly effective in detection and tracking accuracy. Additionally, it has the capability to achieve real-time tracking.