Monitoring Live Parking Availability by Vision-based Vehicular Crowdsensing

The live availability of parking spots is a key enabler of a variety of intelligent parking solutions. A body of research has explored the possibility of using connected vehicles as mobile sensors to measure parking availability. It removes the need for dedicated roadway sensors, achieving wide sensing coverage in a cost-efficient manner. However, the existing crowd-sourcing solutions typically rely on on-board ranging sensors such as radars, lidars and/or sonars to measure locations of the surrounding parked vehicles. In order to lower the barrier for connected vehicles to participate in the crowd-sourced parking availability sensing, this paper investigates a range-free approach to identifying spot-level parking availability. We employ a monocular RGB camera installed in a vehicle and a computer vision-based object tracking mechanism to perceive the surrounding parked vehicles. Analyzing the time series of object tracking results and matching them with a digital map of a parking facility, the system identifies availability of each spot without requiring depth information. The simulation results show that the proposed range-free parking availability sensing system can detect spot-level parking availability with the average accuracy of 83%.