Combine histogram intersection kernel with linear kernel for pedestrian classification

Pedestrian detection is an active research in recent years. Most of the researchers have proposed many methods for detecting pedestrians in static images, especially on how to encode the character of pedestrian images. But fewer attentions are paid on the ensemble of multiple kernels of support vector machines (SVM) on the same pedestrian feature. In this paper, the classification performance of the detector and a multiple kernels combination method, which combines the histogram intersection kernel and linear kernel on histogram of oriented gradient (HOG), is proposed. Experimental results on INRIA human dataset show its efficiency.