On-road vehicle detectioin using histograms of multi-scale orientations

In this paper, we investigated a new feature set, called the histograms of multi-scale orientations (H-MSO), for vehicle representation and detection. The multi-scale orientations on image pixels are calculated using Gabor filters of different scale and orientation parameters. Firstly, we divide the image into cells, and calculate the histograms of multi-scale orientations in each cell by statistics. Then, the values of histogram bins are normalized in each four adjacent cells and are assembled to form the feature set. Finally, the feature set is used to train an SVM classifier for on-road vehicle detection. Experiments validate the proposed feature set and the detection algorithm.