Vehicle Target Detection in Day-Night Mode Based on YOLOv5

With the development of computer technology, autonomous vehicle technology has been rapidly updated. Robust and efficient vehicle detection on the road is the key task of autonomous vehicle environment perception. At this stage, many scholars have done a lot of research on daytime pattern detection using machine learning and deep learning technology. However, there is still a lack of vehicle detection in night mode. The main work of this paper is to create a new data set, and use different models (n, s, m, l, x) of a single-stage target detection algorithm YOLOv5 to verify the effectiveness of detecting vehicles on the road in day and night mode. Three evaluation parameters are used to evaluate the efficiency of the model: recall, precision and mAP. The goal of this paper is to find an efficient model to improve the accuracy of day-night vehicle detection, and provide vehicle condition detection results. This can provide an effective visual perception for the auto drive system.