Study of Human Detection Algorithms using Histogram of Oriented Gradients

Abstract

This work targets on the human detection in static images from the view of computer vision. The interest of such type detectors resides in their many applications like automotive safety, video surveillance, to control the crowd or automatic indexing of image. Detecting humans is a challenging example as humans can adopt various articulated poses, in various backgrounds under the condition of variation in illumination and color. We study and develop a HOG plus SVM based system to find a solution for robust detection as proposed by Dalal & Triggs. The HOG descriptor proposed by Dalal & Triggs turns out to be robust to small changes in the contour of image, location and direction and also for significant variation in illumination and color. HOGs performs equally well for other classes but here we are specifically considering upright humans.