Complex Environment Road Object Detection Algorithm Based on Improved YOLOv5s

A improved YOLOv5s (You Only Look Once 5s) algorithm is intended to dispose of the issues of low detection accuracy, low coverage, and large computational complexity of road object detection algorithms in complex environments. To improve the applicability of the algorithm in various complex scenarios, we draw the CBAM attention mechanism into the CBS module, which is the Backbone benchmark network to enhance the ability to obtain tiny target features and upgrade the recognition rate of tiny targets. By optimizing the CIo U loss function and the alpha IoU loss function, this paper use the alpha IoU loss function to substitute the traditional loss function of YOLOv5s model, this enhancement improves model training accuracy and hasten the convergence rate of the exercise process. The conclusion shows that the improved YOLOv5s algorithm has an accuracy of 94.06%, a recall rate is 92.33%, and an average accuracy is 95.38%. It is superior to other mainstream detection algorithms such as YOLOv5s and SSD, and can achieve road object detection in complex environments. The research results suggest that the improved YOLOv5s algorithm has improved practical efficiency, especially for complex scene object detection, which has important research significance.