Illegally Parked Vehicles Detection Based on Omnidirectional Computer Vision
Yi Tang, Yao yu Chen · 2009
At present, vision-based illegally parked vehicles detection faces a range of issues such as narrowness of detection range, low detection precision and robustness. This paper proposed a technique for illegally parked vehicles detection. Firstly, Omni-Directional Vision Sensors (ODVS) are used to access Omni-directional images of the scene. Secondly, a method based on two backgrounds modeled by Gaussian mixture model (GMM) with different learning rate is presented. Through simple arithmetic, it is capable to segment temporarily static vehicles in the scene. This method is computational efficient and robust because of the avoidance of a series of complex operations of merging, splitting, entering, leaving, occlusion, and correspondence which are met in traditional methodology depending on object-tracking. Thirdly, shadow suppression is used to overcome the impact of vehicles' own shadow on the detection precision. Experimental results show that the technique can effectively detect illegally parked vehicles with high precision and robustness.