The Study Of Pedestrian Detection And Re-identification Method Of Fusion Depth And Color Information

Pedestrian detection and recognition technology is the field of computer vision research hot topic in recent years,pedestrian detection and recognition technology in the intelligent monitoring,criminal investigation,intelligent transportation,motion analysis,play an important role in areas such as unmanned.Among them,the pedestrian detection has a period of history,its technology is increasingly mature,but the test results are also susceptible to light and the environment,the influence of shooting Angle.And pedestrians heavy recognition is put forward in recent years is new topic,heavy pedestrian recognition is mainly refers to judgment to appear in different cameras in different time of pedestrian whether for the same person,it is an important branch in the field of intelligent monitoring research,because from the surveillance camera captured the characteristics of the pedestrian is not obvious,and when the pace of different pedestrian posture,clothing color,will affect the judgments are similar in appearance,so how to improve the accuracy of recognition is a challenging research work.On pedestrian detection and recognition and study difficulties,in this paper,based on the depth of the device information and color information to the research,the research content of this article is as follows:Called depth image and RGB image exist differences in spatial location,this article adopts the way of affine transformation to eliminate the depth image and RGB image in space position error,in order to match the depth of the data obtained with the RGB image data for the next step research.Due to access to the original depth image noise hole problems,This paper deals with the depth image by means of joint bilateral filtering,in order to get real,complete the depth of the image data,the basis for the next step of pedestrian detection.This paper USES the non-maximal suppression(NMS)algorithm,the RandomHough Transform(RHT)algorithm,to obtain the information about the potential location of the human head.Detect the potential location head,and then based on HOG + the SVM algorithm is used to identify the head,head candidate to eliminate detected by mistake,so as to improve the accuracy of detection,and then complete the pedestrian detection research.Pedestrian recognition based on color information of other research mainly divided into two steps: 1.First of all,from the color video image segmentation in the pedestrian,pedestrian images and then divided into several small pieces,respectively to calculate each block HSV color histogram feature,and then all the colors of the sub-block histogram feature fusion together feature vector.2.The distance metric learning:constructed by the method of distance learning appropriate distance function,the original Euclidean distance transform into martensite distance between sample characteristics,and analysis of markov distance sorting pedestrian recognition of other results are obtained.Finally verified by the experiment on pedestrian detection and recognition algorithm proposed in this paper,the experimental results show that the algorithm is effective,this paper adopted effectively improved the pedestrian detection and recognition accuracy.