Illegally Parked Vehicle Detection Based on Haar-Cascade Classifier

Aapan Mutsuddy, Kaushik Deb, Tahmina Khanam, Kang-Hyun Jo

Abstract: Detection of illegally parked vehicles in the urban traffic scene is required for handling unwanted incidents in the roads such as vehicle collision, traffic jam, accidents etc. Due to occlusion, lightning change and other factors the task becomes more laborious. A video-based proposal is presented in this paper to eliminate these problems. Initially, foreground objects are separated using Gaussian Mixture based background subtraction model. Meanwhile, median filtering is used to remove the salt-pepper noise. A simple shadow removal technique based on HSI and YCbCr color model improves the precision of the vehicle detection. Then, edge detection was performed using Canny method for precise tracking of temporarily static vehicles checking centroid co-ordinate values. Finally, for verification of the vehicles a machine learning mechanism known as ‘Haar-Cascade’ classifier is used. However, experimental data shows that proposed system performed well under different conditions with an accuracy of 92%.