Efficient parameter tuning for histogram of oriented gradients

This paper develops an efficient approach of object detection called Histogram of Oriented Gradients (HOG) by taking the power of Self-adaptive Particle Swarm Optimization (SPSO). The HOG indicates locally normalized histogram of gradient orientations features in a dense overlapping grid gives very good results for object detection. The effects of the various HOG parameters overall human detection performance were evaluated; but, the most important difficulties in order to use HOG for object detection generally, is initializing its parameters for special task. The proposed tuning technique is based on finding suitable values for HOG predefined parameters using SPSO. In fact, it selects appropriate values for HOG predefined parameters, not necessarily the best amount. Experimental results show the superiority of this novelty over standard HOG.