Autonomous Driving Perception Method Based on Improved YOLOv8

This study proposes a method based on the improved YOLOv8 algorithm for object detection in unmanned driving environment perception. By introducing the Conv2Former structure and combining the advantages of convolution operation and self-attention mechanism, the model's detection ability for complex backgrounds and multi-scale targets has been improved. The study employed data augmentation techniques such as Gaussian noise addition and random erasure to improve the model's adaptability to noise and occlusion situations. The experimental results show that the improved YOLOv8 algorithm achieves a mAP50 index of 0.462 on the validation set containing occluded samples and low light conditions, which is significantly better than traditional YOLOv8 and other classic object detection algorithms. This study provides new ideas and technical support for the development of autonomous driving technology. In the future, we will further explore deeper network structures, efficient data augmentation methods, and multimodal data fusion to promote the continuous progress of autonomous driving environment perception technology.