An algorithm for road target detection of autonomous vehicles based on improved YOLOv8

With the development of automatic driving technology, the correct driving of autonomous vehicles on the road is a research hotspot at present. Affected by the complex road traffic conditions and driving environment, the current autonomous vehicle's ability to accurately identify road targets through on-board cameras still needs to be improved. Aiming at the problems of near-target error detection and remote target missing detection in road target detection of autonomous vehicles, an improved road target detection algorithm YOLOv8-RTDAV based on YOLOv8n was proposed. To address the issue of false alarms caused by close-range overlapping road targets, we propose a new C2f module, C2f-EFB, which combines an efficient channel attention (ECA) mechanism. To address the issue of missed detections in long-distance road target recognition, we have added a P2 small target detection layer for improvement. To improve the adaptability of the model and better capture image details and feature information, we replaced the SPPF module of the original model with the SPPELAN module and implemented upsampling using the point sampling module DySample. Improve the loss function CIOU of the original model to EIOU to further enhance the detection accuracy of the model for road targets. Experiments have shown that in the KITTI and TT100K datasets, the improved road object detection algorithm has improved accuracy to varying degrees compared to YOLOv8n. Improved the recognition accuracy of close-range overlapping targets on the road and increased the recognition rate of small targets at long distances.