Pedestrians Detection Using Feature Fusion In Static Image

With the rapid development of computer science and technology, intelligent systems are widely used to assist people to live and work, and the machine vision, which is an important part of the intelligent system, has attracted increasingly attention of researchers and enterprises. Pedestrian detection is the basis for many machine vision applications, such as the vehicle safety driver assistance, intelligent video surveillance, virtual reality, and the pedestrian detection in aerial images and rescue disaster victims and other emerging fields in recent years. Currently, the pedestrian detection technology of static image can be divided into two categories:template-based matching method and feature-based classifier approach. The template-based matching method needs to build a large number of pedestrian’s templates, and then to search out the areas with the greatest template-matching degree in the detection images. The effect of this pedestrian detection method largely depends on the quality of the template. Researchers have proposed a number of mannequins for pedestrian detection at present, such as contour models, which are the most widely used, head and shoulders models, and 3D human models. On the contrary, feature-based classifier approach can be described by extracting the image feature information of the pedestrian, then using the extracted features to train a classifier, and employing the trained classifier to detect pedestrians at last. This method, compared with the template-based matching method, has the advantage of low computational complexity, robustness, etc., so this method is more popular in the present study. This paper proposes pedestrian detection algorithms based on a static image multi-feature fusion after analyzing the popular pedestrian detection algorithm. Overall, the main contribution of this paper is as follows:1. To propose a fusion feature HSV-LBP of HSV (Hue-Saturation-Value) color space features and local binary pattern feature (Local Binary Pattern LBP) fuse in pedestrian detection methods. HSV is a global feature, which simply describes the global distribution of colors in an image. However, LBP features can be well described by the local spatial structure of the image, so the algorithm takes into account both the global feature also consider local features, and has the advantages of a less dimension and a higher computing speed. In the MATLAB environment, the results of using Adaboost classifier algorithm to conduct simulation experiments, and comparing the classical histograms of oriented gradients (HOG) feature, LBP features and fusion features of HOG-LBP performance show that the method has high recognition rate.2. Although LBP features can be well described by the local spatial structure information of the image, the complexity of the light shaded pedestrian environment and pedestrian sticking together easily, thereby causing local spatial structure of the image pedestrians have a greater change, which makes the characteristics of LBP poor light noise robustness. As for this shortage of LBP features, the paper proposes a pedestrian detection fusion feature algorithm, which employs color histogram similarity (CHS) in the color space C1C2C3 fused wiith LBP feature. The C1C2C3 color space has relatively stable color invariance, so extracting LBP features in this space can effectively reduce the impact of shadow on the local spatial structure. The simulation experiment among in HOG features, LBP features and HOG-LBP feature in the MATLAB environment shows that the algorithm has high recognition rate.