Real-time vehicle detection algorithm based on a lightweight You-Only-Look-Once (YOLOv5n-L) approach

A vehicle detection algorithm is of great significance for automatic driving technology. Current vehicle detection algorithms suffer from the complex structure, high configuration of hardware requirements, and the difficulty to apply to mobile terminal equipment. In order to solve these issues, this paper proposes an improved YOLOv5 algorithm, named YOLOv5n-L, for lightweight. First, a depthwise separable convolution and a C3Ghost module are used to replace several C3 modules to reduce the model parameters and improve the detection speed. Then a Squeeze-and-Excitation attention mechanism is integrated into backbone network to improve the accuracy of the algorithm and suppress the environmental interference. Finally, a bidirectional feature pyramid network is used for multi-scale feature fusion to enrich feature information and improve the feature extraction ability of the proposed algorithm. The experimental results demonstrate that compared with the original algorithm, the model weight is reduced by 40 % to only 2.3 M. The mean average precision (mAP@0.5) is increased by 1.7 %. The detection speed reaches 80 FPS, which could accurately detect vehicle targets in real-time.