Building a HOG Descriptor Model of Pedestrian Images Using GA and GP Learning

For detecting a pedestrian by using features of images, it is generally needed to establish a reference model that is used to match with input images The support vector machine (SVM) or AdaBoost Cascade method have been generally used to train the reference pedestrian model in the approaches using the histogram of oriented gradients (HOG) as features of the pedestrian model. In this paper, we propose a new approach to match HOG features of input images with reference model and to learn the structure and parameters of the reference model. The Gaussian scoring method proposed in this paper evaluates the degree of feature coincidence with HOG maps divided with angle of the HOG vector. We also propose two approaches for leaning of the reference model: genetic algorithm (GA) based learning and genetic programming (GP) based learning. The GA and GP are used to search the best parameters of the gene and nonlinear function representing feature map of pedestrian model, respectively. We performed experiments to verify the performance of proposed method in terms of accuracy and processing time with INRIA person dataset.