Drone-Based Vacant Parking Space Detection

Abstract

This paper presents a drone-based method for vacant parking space detection using aerial images. Due to the limited field of view, it is better to use a camera mounted on a drone to monitor a huge parking lot. However, a drone-based camera is not fixed to the ground. Thus, there are many challenges to detect parking spaces and classify their status. To detect parking spaces, the RANSAC scheme is first used to estimate the homography relation between the current captured image and the reference parking space. Then, three novel features are extracted from each park space for occupancy condition judgment, i.e., vehicle color feature, local gray-scale variant feature, and corner feature. A deep NN is then trained to determine the occupancy status of each parking space based on the above three features. The performance of our system is evaluated on varies parking lots under different lighting and weather conditions. The average accuracy can be achieved up to 97%.