Research on Target Detection Algorithm for Traffic Scenarios

Aiming at the problems of low model recognition accuracy, easy to miss detection as well as false detection of target detection algorithms in complex traffic environments, this paper proposes an improved target detection algorithm based on YOLO v8. Firstly, SPD-Conv is used to replace the traditional step-size convolution and pooling layer in the YOLOv8 backbone network to reduce the loss of fine-grained information and enrich the feature information of small objects occupying small pixel ratios; then, a small target detection layer with a size of 160x160 is added to the YOLOv8 neck network to obtain a smaller receptive field and solve the problem of inconsistent scales in detecting small targets, which results in the semantic information loss problem. In this paper, experimental validation is carried out on the BDD100K dataset, and the results show that the average accuracy of the improved algorithm proposed in this paper is improved by 6.4% compared with the original YOLOv8, which effectively improves the target detection accuracy of the model for the traffic scenario under the condition of ensuring the detection speed.