Parking lot management application using instance segmentation

Parking lot management application using instance segmentation

In urban areas, it is difficult to find an available parking space. Many detection-based approaches have been proposed for solving the problem. In general, a detection-based approach involves ultrasonic sensors, weight sensors, geomagnetic and infrared detectors. In spite of its effectiveness, such an approach should pay higher costs of construction and maintenance. In this paper, a deep learning-based intelligent parking guidance system (DLPG) is proposed, where the Raspberry Pi system is used to monitor the parking spaces. When a server receives the video stream, the instance segmentation algorithm ＂You Only Look At CoefficienTs (YOLACT) is used to obtain comprehensive information about parking spaces. By experiment, the accuracy rate of the camera set at a real height of 5.4m can achieve 90%, and the overall accuracy is as high as 98.17%. In addition, we discover the important factors in affecting detection results are the pitch angle and the base angle. The most important factors affecting the detection results are the pitch angle and the base angle.