Object detection in autonomous driving scenario using YOLOv8-SimAM: a robust test in different datasets
Zeyang Cheng, Xiaojie Du, Qing‐Yu Liu, Shuguang Zhan, Bo Yu, He Wang, Xiaojun Zhu, Can Xu · Transportmetrica A Transport Science · 2025
With the advancement of deep learning, AI applications in transportation, particularly in target detection and tracking, have made significant strides. However, real-time vehicle detection of autonomous driving scenario still faces challenges of low accuracy and efficiency. This study proposes an improved YOLOv8-SimAM network, aiming to enhance detection accuracy. The SimAM attention module is introduced and the loss function is optimised to enhance the features of target vehicle without increasing network parameters. YOLOv8-SimAM is combined with DeepSORT for efficient tracking. Experiments show that the YOLOv8-SimAM improves mAP by 1%, accuracy by 4%, and FPS by 3.36% on the UA-DETRAC dataset compared to YOLOv8. It also performs excellently on BDD100k and MIT vehicle datasets, addressing multi-target detection issues in complex scenarios. The model’s improved accuracy and efficiency, along with its strong generalisability and reliability, make it suitable for various driving environments, providing robust support for full-scale autonomous driving.