Research And Design On The Algorithm Of Pedestrian Detection Under Infrared Image Technology

Pedestrian detection based on computer vision has become one of the most active research topics of computer vision and intelligent vehicles’field because of its important applications values in Driver Assistance. The purpose is to detect pedestrian by installed cameras in the moving vehicles, and estimate the potential risk to take the strategy to protect pedestrians.This article first discusses the general concept of pedestrian detection, and some of the current detection techniques, then make improve of the detection algorithm under infrared images. In detection algorithm of infrared image, the first step is segment the region of interest in the image, which is the most critical step of pedestrian detection algorithm.This article focuses on the design of the pedestrian segmentation. First, conducts the grayscale conversion for infrared image, then removes the noise of image and enhances the image. Considering the single-threshold binarization processing results are not an ideal situation, this article takes a dual-threshold segmentation method, and by the study of genetic algorithm, this article propose the interactive genetic algorithm applied to the field of image preprocessing, namely the use of interactive genetic algorithms to find optimal parameters of dual-threshold binary image for achieving better results of image segmentation. After getting a better binary image, uses morphological operations and image processing to remove a small area. In the step of segmentation of region of interested, this article proposes a algorithm which components of external rectangle and symmetry characteristics to extract the candidate regions of potential pedestrian.In the part of pedestrian recognition, extract the HOG feature of candidate region of the pedestrian. Inputting the library of pedestrian and non-pedestrian to the SVM support vector machine for offline training, after obtaining the classification function, it can identified if the area is a pedestrian after inputting HOG features.In the part of simulation, this article first discusses the influences of different ways to find the optimal parameters of the image segmentation. Then makes experiment under road scene using visual Studio2005and the Matlab technology, and analysis the results of simulation. Experimental results show that the algorithm is reasonable and effective.