Traffic light recognition and detection of unmanned vehicle based on improved YOLOv5

To address the issues of large model size and slow detection speed in traffic light detection for autonomous vehicle network models, a lightweight object detection algorithm based on YOLOv5 is proposed. First, the Ghost Net network is used to replace the conventional convolutional network (Conv) in the traditional YOLOv5 algorithm to enhance the nonlinear approximation capability of the network model. Second, the large-scale object detection feature of the traditional YOLOv5 model is removed, retaining only the medium and small-scale object detection, reducing the model size and speeding up traffic light detection to achieve lightweight performance. Finally, training and experiments were conducted on a custom dataset, and the results showed that, compared to the traditional YOLOv5 model, the improved model reduced its size by 44.6% and increased its speed by 12.6%. These improvements make the application of the YOLOv5 algorithm more feasible in real street scenarios.