Improvement of Feature Extraction Based on HOG

In recent decades, with the rapid development of science and technology, pedestrian detection has gradually begun to mature from the beginning. Pedestrian detection involves a number of disciplines and fields to achieve joint cooperation. As the basis of pedestrian detection, image processing needs to ensure the quality and speed of detection at the same time. Face recognition based on directional gradient histogram (HOG) has good accuracy in pedestrian detection. But at the same time, compared with other pedestrian detection feature extraction methods, the disadvantage of hog is that it takes too much time and can not guarantee the detection speed while improving the accuracy. On this premise, based on the idea of clustering, the hog features are clustered according to their gradient directions, and then the appropriate features are found out by statistical calculation to form the combined features, and the subsequent steps are carried out by the combined features. Through the experiment, without sacrificing the detection accuracy, the detection efficiency can be effectively improved by reducing the data dimension.