Dense Object Detection Algorithm in Traffic Scenes Based on Improved YOLOv5s
The object detection tasks in road traffic scenarios often involve blocking each other and the pixels of distant objects are low,resulting in low object detection accuracy,missed detection,false detection and other problems.Therefore,an improved YOLOv5-ECFK model for intensive object detection in traffic scenarios is proposed.Firstly,the boundary box regression loss function is replaced by CIOU with EIOU to improve the accuracy of the model evaluation prediction frame.Secondly,CBAM attention mechanism is integrated into the main network structure to help the network find the attention region in the scene with dense objects.Then,the detection accuracy of the model is improved by adding the small-scale feature fusion layer and detection layer.Finally,the K-Means++algorithm is used to re-cluster the anchor frame,so that the generated anchor frame is more suitable for the experimental data set,and the overall performance of the algorithm is improved.The experimental data set is composed of the open data set Rope3D and the self-made data set.Experimental results show that compared with YOLOv5s,mAP@0.5/％and mAP@0.5:0.95/％,the final improved algorithm YOLOv5-ECFK are improved by 3.4％and 2.7％,respectively,and the Precision and Recall are improved by 3.5％and 1.3％,respectively.Finally,the improved algorithm has excellent detection effect on dense objects in road traffic scenarios.
万方数据
维普
