The Pedestrian Detection Algorithm Based On The Texture Features Of Wavelet Transform And Secondary Authentication Of Multiple Components

Pedestrian detection technology is one of the hottest areas of machine vision research and has been widely used in the intelligent auxiliary driving, intelligent robot, human behavior recognition, etc. A series of pedestrian detection algorithm based on statistics has been proposed after unremitting research in recent years which can be divided into two parts: extracting the features from samples and the design of classifier. The purpose of the sample’s feature extraction is to extract the most representative characteristics of samples. The design of classifier belongs to the field of machine learning which is to get more effective, more quickly and better generalization of the classifier.Based on the analysis of current situation and bottlenecks of domestic and foreign research about pedestrian detection, this paper proposed to improve the current main methods of feature extraction. The main contributions of the paper are as follows:1、Discrete wavelet transform of Haar-LL pedestrian detection researchTo solve the problem of the low detection accuracy and high detection time consumption of pedestrian detection, a two-dimensional discrete wavelet transform Haar Local Binary Pattern (LBP) combined with Local gradient Pattern (LGP) feature fusion method (Haar-LL) was proposed. Firstly, the two-dimensional discrete Haar wavelet transform was applied to image and four different frequency sub-images was obtained. Then, extracting LBP feature from the low frequency sub-image and LGP feature from the three high-frequency sub-images respectively. Finally, fusing the three characteristics of LGP parallel and LBP features serial. Comprehensive various experimental data show that this method is feasible and effective.2、Multi-component secondary verification pedestrian detection algorithmAiming at the solving the shortcomings of missed detection of occluded pedestrian, feature inhibition caused by feature fusion and expensive computation cost of pedestrian detection algorithm, this paper presents a multi-component secondary verification pedestrian detection algorithm. A first pedestrian detection can be made using the fusion feature of LBP feature extracted from whole image and LGP feature extracted from the LL low frequency image of the 2-D discrete wavelet transform. If the decision value of detect result is larger than the threshold, the program are stopped; otherwise LBP feature of head and shoulder and HOG feature of legs are extracted to secondary verification using body parts to refine detect results. Experiments show that that the method has little side effect to the detection efficiency but improves the detection accuracy rate.