Pedestrian Detection Using Histograms of Oriented Gradients (HOG)

Pedestrian detection is one of the most fundamental aspects of computer vision.

To understand how a Histogram of Oriented Gradients (HOG) works, we must first know what a gradient and a histogram are.

The gradient measures how rapidly and in which direction the level varies in a black-and-white image. The grayscale level is similar to height in a black-and-white image. A substantial gradient from white to black perpendicular to a picture edge (a black-to-white transition) results from a strong gradient perpendicular to the edge.

For each pixel, the system computes a gradient, and the gradients are used to fill a histogram: the value is the gradient’s angle, and the weight is the gradient’s magnitude. To determine whether the cells in the current detection window match a person or not, the system combines the histograms of all cells and sends them to a machine learning discriminator.

This technique has been designed to identify pedestrians who are completely visible and standing up. As a result, we should not expect it to function in other situations.