Muthanna Journal of Engineering and Technology
Volume 5 issue 3, 2017
Dr. Ahmed Freidoon Fadhil
College of Engineering, University of Kirkuk
The visual analysis of moving objects has been an important computer vision research area. As the number of vehicles is increasing on the roads, the need for accurate detection of vehicles is rising. A hybrid segmentation method that combines background subtraction, threshold segmentation, morphological operators, and watershed segmentation is proposed in this paper.
The shadow has a vital effect on the performance of many fields like tracking, classification, detection, and shape analysis. The shadow presented in the data used in this paper was successfully removed using watershed and shape analysis. Since the connected vehicles touch each other in the boundary only, the watershed transform can correctly isolate these touching cars. The proposed system overcomes the over-segmentation drawbacks of the watershed transform by applying it to the gradient of the image rather than the image itself.
Finally, shape analysis is used to remove large shadow parts which lead to the detection of the vehicles only. An accuracy detection rate of 97% was reported from the highway video which is astonishing result compared to existing methods. The proposed algorithm was coded using MATLAB R2014b programming language. The accuracy and simplicity of the proposed system make it applicable for real-time traffic surveillance Applications.