Research on Lightweight Adaptive Object Tracking Algorithm based on Improved YOLOv11n and DeepSORT

Abstract

To address the issues of complex model structure, poor real-time performance, and susceptibility to environmental interference in two-stage object tracking algorithms, we propose a lightweight adaptive object tracking algorithm based on an improved YOLOv11n and DeepSORT. The YOLOv11 algorithm is enhanced by incorporating the RepNCSPELAN4-low module to reduce model parameters; the C3k2 module is used to replace the RepC3 module in the CCFM network structure to improve feature expression and further reduce both the parameter and computational loads. Additionally, the SENet attention mechanism is introduced to enhance the network’s ability to extract global contextual features. For the tracking part, the DeepSORT algorithm is combined with the ByteTrack framework to improve trajectory matching accuracy; adaptive Kalman filtering is applied to optimize the motion model; and the CIOU matching mechanism is introduced to enhance the stability and accuracy of target association. Experimental results show that the improved algorithm, compared to the original YOLOv11, increases FPS by 26.1%, reduces computational load by 9.4%, decreases the number of parameters by 35.3%, and lowers the model size by 32.7%. Furthermore, mAP50 and mAP95 reach 86.8% and 62.5%, respectively. The enhanced object tracking algorithm achieves a 1.9% improvement in MOTA, a 5.5% increase in HOTA, and a 5.3% rise in IDF1, with an FPS of 27, meeting the requirements of practical applications.