Car parking detection in images by using a semi-supervised modified YOLOv5 model

The problem of car parking detection in images attracts the attention of many researchers. In this task, it is quite difficult to identify rectangular, continuous parking spaces in all kinds of city images under different weather conditions, combining the low-light environment and the system’s low cost with high detection accuracy. In this paper, we propose a modified version of the YOLOv5 model joined with semi-supervised learning that allows us to detect parking lots in any complex scene, independent of parking space lines and parking environments. Due to the combination of the nature of semi-supervised learning and the high accuracy of supervised learning models, the modified version of YOLOv5 model permits to use very little labeled data and a large amount of unlabeled data. It can significantly reduce training time while maintaining recognition accuracy. Compared with other neural network models, the modified version of YOLOv5 model has the characteristics of fast training speed, persistent operation, small model size, and high model precision and recall values.