Research On Pedestrian Recognition And Tracking Ahead Of Vehicle Based On Infrared Images

Pedestrian recognition and tracking ahead of vehicle based on infrared images is a key technology in driver assistance systems,which has become one of the hot topics in the current field of vehicle active safety. An infrared thermal imager installed in the front of the vehicle,it is used to get pedestrian information. As infrared has the ability to penetrate clouds、fog、rain、snow furthermore, it is not affected by glare or insufficient light conditions, the infrared sensors can be used real time in all weather conditions or day and night. Especially,the infrared thermography assists all drivers at night,which helps reduce the incidence of accidents and protects pedestrian’s life and property safety.This paper treat infrared images as the object of study, we mainly research on identifying pedestrian target and locating the position of pedestrian in infrared image, and solves the single moving pedestrian tracking problem of occlusion in complex environment.This paper firstly illustrated the principles of infrared imaging,we determined a certain method to preprocess infrared image by analyzing the features and advantages of the infrared image.In this paper,we adapt the median filter and power transform to improve the signal-to-noise ratio and contrast of infrared image.We choose the the Otsu threshold segmentation, this method can effectively segment the regions of interest from image.Classificating the pedestrian target and locating the position of pedestrian in infrared image.In this paper,we trained two classifier using shape cues and Haar feature of human body.Utilizing the support vector machine (SVM) classifier to eliminate non-pedestrian object,and then exploiting the Haar feature classifier to help locate the exact position of pedestrians. Experimental results show that the two classifiers can solve classification problems of pedestrian, and locate the exact position of pedestrians with red rectangular box.According to mean shift algorithm is applied to the infrared image.and it is unable to cope with the occlusion problem, as well as the characteristics of the Kalman filter tracking algorithm,which can predict the position of moving target.I proposed a algorithm that combine mean shift algorithm with Kalman filtering.This paper use the Kalman filter to predict target position,and then, mean shift algorithm search the accurate position around the predict area.The proposed algorithm effectively solved the problem of occlusion in target tracking.