A multi-object detection and tracking algorithm for autonomous driving based on improved YOLO26 and OCSORT
With the rapid advancement of autonomous driving technology, multi-object detection and tracking has become a critical task in environment perception. To address the challenges of large-scale variation, severe occlusion, strong background interference, and frequent identity switching in complex traffic scenarios, this study proposes a multi-object detection and tracking method based on an improved YOLO26 and OCSORT framework. In the detection stage, the YOLO26 network architecture is optimized by incorporating a more powerful feature extraction module, a multi-scale feature fusion strategy, and an attention mechanism, thereby improving the accuracy, robustness, and real-time performance for detecting traffic participants such as vehicles and pedestrians. In the tracking stage, OCSORT is further enhanced through improvements in motion state estimation and data association, which strengthens the continuity of tracking under occlusion, crossing, and rapid motion conditions, while reducing target loss and identity switches. Experimental results demonstrate that the proposed method achieves improvements in mAP50:95 of 5.60% and 3.40%, and improvements in multiple object tracking accuracy of 3.31% and 2.90% on the KITTI and BDD100K datasets, respectively. Overall, the proposed method outperforms other approaches in terms of accuracy, stability, and real-time performance, and can provide effective technical support for multi-object perception in autonomous driving scenarios.