HOG is a feature descriptor used in object detection. It works by calculating the distribution (histogram) of gradient orientations in local regions of an image. HOG captures the shape and edge information of objects, which is useful for tasks like pedestrian detection in surveillance videos. By dividing the image into small cells and computing the gradient histograms for each cell, HOG can represent the object's appearance in a way that is relatively invariant to small changes in pose and illumination. It is widely used in computer-vision-based security systems and autonomous vehicle navigation for detecting obstacles.
