Research On Target Recognition And Tracking Based On Stereo Vision

The target identification and tracking technology have a wide application in many fields,and gradually become automatic and intelligent.In this paper,the binocular camera that can obtain spatial three-dimensional information is used to realize the object recognition and tracking research.The binocular camera has the characteristics of simple image acquisition,high speed,high efficiency,low cost,and the ability to construct a three-dimensional space.In this paper,the binocular camera is used to collect the target image,identify the target and determine the position of the target based on extracting and matching the feature points of the object image and the video stream image.The attitude determination and dynamic tracking of the target image in the video stream image are completed by the principle of binocular camera imaging and coordinate system conversion.This paper mainly contains the following contents:Firstly,the problem of target recognition and location in binocular camera image is solved.In the binocular camera image,local feature points are extracted and matched from the known image and the camera image.The matching experiment between the known image and the unknown scene image is carried out by using SURF algorithm.Then the optimal matching point pairs are selected and the two-level optimization of these is carried out by using RANSAC algorithm,where homography matrix of the object region can be obtained.The homography matrix is used to derive the position of the object in the image.Secondly,a horizontal relative pose recognition algorithm is proposed based on the principle of image difference in binocular images of space objects.According to the internal and external parameters of the camera obtained in the calibration,the coordinate difference between the left and right camera images calculate the angle on account of the conversion relationship between the imaging principle of the binocular camera and the coordinate system.The horizontal angular attitude of the target object corresponding to the binocular camera is derived by the corresponding triangular geometric transformation.Finally,the validity of the horizontal attitude estimation method is verified by experiments.The feasibility of combining SURF target recognition with attitude determination algorithm is discussed and verified.Finally,the dynamic tracking problem of the target object in the camera image is studied.The proposed algorithm,which introduces the position contour information of the target object in the image to be matched by the SURF algorithm into the tracking algorithm KCF,cancels the step of manually calibrating the target area by the KCF algorithm,simplifies theparticipation of the human and liberates the human hands.Finally,through the specific target and camera,the algorithm is experimentally verified.From the experiment,it can be seen that the target object in the video stream can be identified and tracking automatically and intelligently.