K S Chidanand Kumar, National Institute of Technology, India

This paper presents a novel approach for detecting vehicles for driver assistance. Assuming flat roads, vanishing point is first estimated using Hough transform space to reduce the computational complexity. Localization of vehicles is carried using horizontal projection on the horizontal gradient image below vanishing point. An uppermost and lowermost peak in the horizontal profile corresponds to search space of vehicles. Binarization of search space on the horizontal gradient image is done using Otsu algorithm. Verification of vehicles is carried through a series of rule based classifiers constructed using statistical moments, observing peaks in vertical profiling, vehicle texture, symmetry and shadow property. Experimentation was carried out on flat highway roads and detection rate of vehicles is nearly found to be 88.23%.

Vanishing point, Horizontal projection, Vertical projection, Otsu, Gray Level Co-Occurrence Matrix, Entropy, Homogeneity, Vehicle Gray Scale Symmetry.