Vision-Based Irregular Car Parking Behaviors Detection in the Underground Garage

Irregular parking behaviors in the underground garage are troubling car drivers and property. To solve the problem, this paper proposes a vision-based detection method of irregular parking behavior in the underground garage. The irregular parking behaviors, which include parking with the tail facing out, parking in other people’s parking spaces, parking in no-parking areas, and not parking in the parking lines, can be auto-detected depending on the on-site inspection of an automatic patrol robot that embeds the proposed detection method. First, YOLOv3 is used to train the object detection model, which is divided into the headstock, tailstock, side of the car, license plate, and parking line areas. Second, object detection and classification are obtained based on a certain number of frames captured by the robot camera. The position information of the vehicle is obtained by scanning the QR codes, and the corner points of the parking line are detected. Finally, the irregular parking behaviors of vehicles are detected by the judgment algorithm, and relevant information is recorded.