Research And Implementation Of Human Body Image Detection Based On HOG

Human detection has many applications in computer vision, such as video tracking, smart cars, human detection, intelligent robot and so on. An automatic method for finding humans in a scene serves as the first important preprocessing step in understanding human activity. The challenges are due to a wide range of poses that humans can adopt, large variations in clothing, as well as clustering background and complex illumination.At present, in the field of human image detection, binary sliding window is used to classify and detect. The step of this method is exhaustive scanning different size images with fixed or variable size detection window in use of specified step size and trajectory. In the detection, the feature information of each detection window is input classifier for training, then classify the image with trained classifier. Features can be used for human detection are in large quantities, including HOG feature, i.e. gradient direction histogram feature, poposed by Dalai and Triggs in 2005, which has excellent performance.In this paper, firstly according to HOG features' characteristics of high accuracy and large amount of calculation, Multi HOG feature which is structured by Fisher selection criteria is selected instead of traditional HOG by adjusting the structure of HOG, and the multi scale blocks are used to replace the original blocks, result in the feature dimension reducing to 360 dimensions form 3780 dimensions in the traditional HOG. Secondly, to further enhance the detection results, LBP feature is coalesced on the basis of Multi HOG features, which has good performance in texture.In this paper, the nonlinear kernel SVM is used instead of the linear SVM because of the poor classification effect of the linear SVM and the purpose of reducing the detection time, improving the detection efficiency and finally achieving the pedestrian sliding window detection.In this paper, the INRIA standard data set is used to test the human body detection method. The experimental result shows that the proposed method can complete human detection task well, and the detection results are better than the original HOG features, and the proposed method has certain advantages in the detection time.