Localized Background Subtraction Feature-Based Approach for Vehicle Counting
Mohamed A. El-Khoreby, S. A. R. Abu–Bakar, Musa Mohd Mokji, Syaril Nizam Omar, Najeeb Ur Rehman Malik · 2019
This paper presents vehicle counting using local features derived from background subtraction method combined with the traditional SVM classier. The main advantage of the proposed method is that it manages to overcome counting problem in different challenging situations such as during rainy day condition, which is yet a difficult problem in the standard background subtraction method. In this method, the background subtraction is performed within a small predefined region of a lane or road, thus increasing the computational speed. Next, two key features i.e. area covered by the motion indexed graph and number of edge pixel computed in the designated areas in the image are computed to form a 2D feature vector. Finally, we use SVM for classifying this vector into either vehicle or noise. Comparisons between the proposed method and other method show the potential of our approach.