A popular feature descriptor for object detection is the Histogram of Oriented Gradients (HOG). HOG descriptors can be computed from an image by first computing the horizontal and vertical gradient images, then computing the gradient histograms and normalizing across blocks, and finally flattening into a feature descriptor vector. These normalized block descriptors finally obtained are called HOG descriptors, a feature descriptor used in a variety of computer vision and image processing applications for object detection.
The use of HOG descriptors has been particularly successful for detecting humans, animals, faces, and text. We already described how to compute a HOG descriptor from an image. At first, a (linear) Support Vector Machine (SVM) binary classifier model is trained with several positive and negative training example images. Positive images...