Vehicle and Pedestrian Detection Algorithm in an Autonomous Driving Scene Based on Improved YOLOv8

Aiming at the problems of insufficient lightweight and slow running speed of vehicle and pedestrian detection model in autonomous driving scenarios, a vehicle and pedestrian detection algorithm based on improved YOLOv8 (You Only Look Once version 8) was proposed to realize an intelligent, safe, and efficient autonomous driving system. First, FasterBlock in FasterNet replaces the Bottleneck in C2f, which reduces the number of parameters in the model and improves the model’s real-time detection performance. Second, the EMA attention mechanism is used to fuse with C2f-Faster to improve the feature fusion ability of the model, and a new C2f-Faster-EMA module is designed to replace part of C2f. Then, the object detection head Dynamic Head based on the attention mechanism is introduced, and the deformable convolutional DCNV3 is used to replace the deformable convolutional DCNV2 in Dynamic Head. A new Dyhead-DCNV3 module is designed to replace the original detection head Detect. Finally, the ablation experiment verifies the function of each module of the improved model, and each module’s contribution to the target detection performance is analyzed. Experimental results show that compared with the original model, the mAP of the improved model in the customized automatic driving scene data set is increased by 1.4%; the parameters of the model are reduced by 5.7%; and the running speed of the model is up to 178.6 FPS, which is very competitive with other algorithms.