With people's increasing requirements for quality of life, intelligent vehicles have become an indispensable means of transportation in daily life. However, in dense roads, the accuracy of vehicle detection is not high enough. The objective of this research endeavor is to enhance both the precision and the speed of vehicle target detection. In view of the limitations of traditional YOLOv8 vehicle detection in dense roads, a series of innovative improvements are proposed. Firstly, we introduced CBAM (Convolutional Block Attention Module) attention mechanism, which effectively focuses on the key features of the vehicle and enhances the performance of the model to capture important information. Secondly, the C2f_DCN module is used to improve the richness and accuracy of feature extraction. Finally, the WIoU loss function was selected to optimize the accuracy of bounding box regression. Through experiments on real-world vehicle datasets, the precision of the improved YOLOv8s model is increased by 10.9%, the recall rate is increased by 3.76%, and mean average accuracy(mAP) is increased by 2.89%, which offers a dependable and streamlined approach to vehicle detection within intelligent transportation systems, enhancing their overall performance.
Vehicle detection research based on improved YOLOv8s
Fifth International Conference on Optical Imaging and Image Processing (ICOIP 2025) ; 2025 ; Xi’an, China
Proc. SPIE ; 13688 ; 1368803
2025-07-28
Conference paper
Electronic Resource
English
Deep Learning Network Model for Road Pavement Damage Detection Based on YOLOv8s-GES
Transportation Research Record | 2025
|