Improvement of multi-object detection on highway based on YOLOv11

In highway object detection, the challenge lies in how to accurately and real-time locate and classify various targets in images, such as sedan cars, coaches, construction vehicles, construction workers, and so on. This capability is crucial for identifying abnormal behaviors and events on highways. To more effectively meet these requirements, a detection algorithm based on YOLOv11 is proposed to further improve detection accuracy. Firstly, to enhance spatial resolution and retain more details of small objects, the P5 layer is replaced by the P2 layer. Secondly, a Unified-IoU (UIoU) loss function is introduced. UIoU automatically assigns varying levels of importance to prediction boxes at different quality levels. This improves object detection accuracy, and the probability of false positives and missed detections is successfully reduced. To validate the performance of the proposed algorithm, we conducted comparative and ablation experiments. When the two modules are applied together, the improved model’s mAP50increases by 3.5% compared to the baseline, and mAP50−95increases by 1.5%. Furthermore, the method maintains competitive real-time performance.