This paper addresses the challenges of vehicle occlusion and insuffi-cient real-time identification in the context of target recognition for autonomous driving scenarios. It proposes a lightweight improved target recognition algo-rithm based on YOLOv8n, named MMW-YOLO. The algorithm employs a more streamlined MobileViTv3 structure, significantly reducing the number of model parameters while maintaining accuracy. Furthermore, it embeds the Multi-scale convolutional attention (MSCA) module to enhance recognition accuracy. Lastly, the Wise-IoU loss function is utilized to boost the model’s generalization. Experiments conducted using the KITTI dataset demonstrate that MMW-YOLO achieves a 5.2% improvement in average precision over the original model and a 53% reduction in parameter count. The experiments verify the superior performance of the MMW-YOLO algorithm for target recognition tasks in autonomous driving scenarios.