Using Computer Vision Techniques for Parking Space Detection in Aerial Imagery

We propose a pattern recognition algorithm that uses computer vision to analyze and map the location of parking spaces from aerial images of parking lots. The analysis method developed made use of line detection coupled with selective filtering based on the prevalence of line length and angle. The goal of this algorithm was to provide a means of automated detection of regions of interest in parking lot images for further use in collection of parking data. Aerial images that used for development and testing were collected via a quad-copter. The quad-copter was equipped with a camera mounted via a gimbal that maintained a camera angle parallel to the parking lot surface. Video was collected from an altitude of 400 ft and individual frames were selected for content. The images were then split into a development set and a testing set. For analysis, images were converted to gray-scale followed by the application of a binary filter. Line features in the binary images were then detected using a Hough transform. Resulting features were then analyzed iteratively to find recurring line patterns of similar length and angle. After filtering for noise, line end-points and intersections were grouped to estimate individual parking space locations. We performed analysis of the proposed methodology over real spaces and demonstrate good performance results.