A Computer Vision-Based System to Study Parking Utilization

Parking studies are commonly used to evaluate the use of parking lots by determining peak usage days and times, balance of lots usage, and similar other aspects. These studies have traditionally depended on human labor with researchers walking through a series of parking lots at scheduled times to record the number of occupied parking spaces. The recent availability of camera-equipped unmanned aerial vehicles (aka drones), together with the advancements in computer vision and deep learning techniques made it possible to replace the human labor-based approach with a more efficient, cost-effective, high accuracy automated approach.In this paper, we present an end-to-end computer vision-based solution that was developed and tested in parking utilization studies on our university campus. A drone, programmed to travel on a flight path over the parking lots of interest, captures images at different times. The images are processed and fed into a pre-trained deep learning model that detects the locations of vehicles in the images. A Python script counts and reports the number of vehicles parked in the various sections in each lot at every time. The results show that the system was able to successfully report the number of vehicles with a 100% accuracy rate.