In Intelligent assisted driving scenarios, vehicle detection may have small targets and severe occlusion, resulting in incomplete perception of the traffic environment. To solve the above issues, a vehicle detection method based on improved YOLOv7 is proposed to prevent information loss during feature transmission. Firstly, the channel and spatial attention branch of BAM attention mechanism are improved to reduce the loss of feature information; The meta-ACON activation functions is used to enhance the effective expression of features in the network; Finally, weighted feature fusion and cross layer connectivity are introduced to fuse features from different layers to further avoid information loss. The experimental results show that the mAP of the improved method on the KITTI vehicle dataset is 96.05%. Compared to the original YOLOv7, the detection speed is basically unchanged, and the accuracy is improved by 3.62%. Moreover, the detection effect for small target vehicles and occluded vehicles is significantly improved.