Infrared multi-target tracking based on Deep-Sort optimization algorithm

• Published in: International Conference on Control, Automation and Information Sciences (Online) pp. 1023 - 1028
• Main Authors: Ling, Feng, Zhang, Yan, Zhang, Jinghua, Shi, Zhiguang, Zhang, Yifan, Suo, Yuchang, Liu, Xinpu, Wang, Fang
• Format: Conference Proceeding
• Language: English
• Published: IEEE 14.10.2021
• Subjects: Adaptive optics; background matching; DeepSort; Heuristic algorithms; Indexes; IOU; Kalman Filter; Optical filters; Optical imaging; Solid modeling; SSIM; Target tracking; Yolov5
• ISSN: 2475-7896
• DOI: 10.1109/ICCAIS52680.2021.9624607

This Multi-target tracking has always been an important topic in the field of computer vision. Due to the poor detection and tracking accuracy of traditional multi-target detection algorithms, it cannot meet the requirements of high-precision tracking tasks in a complex background environment and diversified target scenes. In order to realize the intelligent perception and recognition of infrared optical images under airborne conditions, this paper adopts the detection-tracking method to carry out research on the subject based on the latest achievements of deep learning. In the infrared optical target detection part, the rapid detection method of infrared optical target based on Yolov5 is adopted, and the closed-loop processing mode of "task modeling - scheme realization -result evaluation" in engineering application is used for reference. Through quantitative evaluation of network output detection results, dynamic and benign adjustment of data is made. In the stage of infrared target tracking, the infrared optical multi-target tracking algorithm based on Deep-Sort is adopted. On the basis of the algorithm, the SSIM structural index, background matching algorithm and target IOU feature are used to modify the Deep-Sort tracking trajectory, and the missed detection rate and false detection rate of the target are optimized.