Vehicle Recognition under Autonomous Driving Based on YOLOv5

In the field of autonomous driving, a key requirement is the fast and accurate identification of vehicles on the road. While many researchers have conducted extensive research on vehicle detection in general road environments using deep learning techniques, there are still shortcomings in vehicle detection under certain special driving conditions. In this study, the manually processed datasets were used to train and test the YOLOv5s model, which is based on a single-stage object recognition algorithm. The influence of the datasets on the recognition performance of the model was confirmed by the test results. The recognition model was trained twice. In the first training, the specially processed datasets were not included in the training process, resulting in a final accuracy of 89%. In the second training, the processed datasets were included in the training process, resulting in a lower accuracy of 87%, and the second trained model was able to quickly and accurately identify vehicles even in the presence of noise interference or incomplete vehicle images.