Abstract: Investigation and punishment of vehicle parking violations is an important part of urban traffic management. Considering of the time-consuming and labor-intensive nature of manual law enforcement, as well as the limited scope of fixed camera monitoring and detecting, exploring more flexible and efficient automatic detection methods has great practical significance. Thus, a cruise detection technology suitable for road moving carriers that does not require stopping and can be completed in one go was proposed for this purpose. Firstly, a vehicle parking violation image dataset named XMUT-VPI was collected and constructed under the conditions of approximate horizontal views and random shooting angles, laying a data foundation for the research. Then, a multitask neural network MTPN was constructed to simultaneously carried out vehicle target detection, tire contact location, and parking line segmentation. By help of the self-designed deformable large kernel feature aggregation module and cross-task interaction attention mechanism, an highest average detection accuracy of 90.3%, an minimum average positioning error of 4.4%, and an suboptimal average segmentation intersection ratio accuracy of 78.5% were achieved. At last, based on the network inference results, skeleton as well as it’s features of the parking line were extracted and visible area of the main parking space was fitted. These help to match a target vehicle and analyze the positional condition between its tire groundtouching points and the main parking space. Based on all these works, a judgment algorithm was provided for three typical behaviors: illegal parking, improper parking, and standardized parking. Experimental results confirmed that the algorithm attains a comprehensive accuracy rate of 98.1% for vehicle parking violation detections across diverse complex interference scenarios, which leads existing mainstream methods and is able to provide technical supports for fully automatic road cruise management.

Key words: deep learning, vehicle parking violation, object detection, key point localization, semantic segmentation