Real-Time Parking Occupancy Detection for Gas Stations Based on Haar-AdaBoosting and CNN

Abstract

Urban parking management is receiving significant attention because of the explosive increase in the number of vehicles. This paper proposes a novel parking occupancy detection method based on a Haar-AdaBoosting algorithm and a convolutional neural network (CNN), which provides intelligent video surveillance in public parking areas, especially gas stations. A key step in this method is accurate, real-time detection of vehicles. An initial set of sub-windows, which contain possible vehicle regions, are obtained by the Haar-AdaBoosting cascade classifier, and these sub-windows are passed to the CNN to filter out non-vehicle regions. Then, we propose a method of parking occupancy detection to recognize which parking spaces are occupied. This method can also recognize illegal occupancy. Our experimental results show that the proposed vision-based method of parking occupancy detection for gas stations is capable of suppressing false positive and can recognize parking occupancy effectively.