The Research And Realization Of Intelligent Vehicle Tracking System Based On Machine Vision

Machine vision is an important scientific field that uses the machine in place of human eyes to detect, identify and track target, which is mainly used in large quantities of industrial production, drug testing, bank card demo system, surface inspection and automobile electronic industry. Since the 1950s, automotive electronics and intelligent control technology have been appreciated extensively and have developed rapidly.Subsequently, intelligent vehicle then becomes a hot research in the field of automotive control, and vision tracking also becomes one of the main directions of intelligent vehicle research. The identification and track for moving target in intelligent car is to separate the object from image sequences which are obtained from visual sensor and to track and identify the object the background, therefore, the target detection, identification and tracking is an important mission of intelligent car vision tracking.The thesis designs an intelligent car tracking system based on high-speed spherical camera, which tracks two cars in the view of camera. PC unit gets the motion state of the car in front of one smart car through the camera and and uses car machine vision algorithm to control the rear intelligent car to track target car through wireless command. During the tracking, the information about the objective location is transmitted by serial communication to control the PTZ in order to track the two objects to ensure them being inside of scene all along. For this system, the research mainly includes three parts: first, moving target detection and extraction; second, moving target tracking; third, the speed and direction of camera and intelligent vehicle control.Several commonly motion detection algorithms are firstly introduced, including optical flow, time-difference, background-difference, and the sphere of applications of these algorithms have been verified and analyzed. The method of foreground detection based on the mixture of Gaussian model is emphasized especially. Then the basic principle of MeanShift algorithm is recommended, and CAMShift target tracking algorithm, Kalman motion estimation tracking algorithm and particle filter tracking algorithm is thoroughly studied. On this basis, the paper has been fulfilled following works:(1) A system combining mixture of Gaussian background model with CAMShift algorithm has been designed to detect and track automatically the moving target. For detecting the moving object, the region which the movement object belongs to is firstly identified and extracted by mixture model of Gaussian, and the centroid of this region is determined as the center of initializing window for tracking. The color feature of object is then extracted in the region, and the CAMShift algorithm is used to calculate the exact location of the target and adjust the size of search window. During the tracking, the information about the objective location is transmitted by serial communication to control the PTZ in order to track the object to ensure it being inside of scene all along.(2) On the basis of above algorithm, some improvements to CAMShift are made. First, color features and the texture characteristics of target is fused.HSV color histogram and the gradient tonal histogram will be weighted as histogram target features templates to deal with similar color interference; Second, adopt adaptive extended search window to improve CAMShift algorithm and avoid missing target for the excessive instantaneous velocity; Third, combine Kalman filtering with CAMShift algorithm to predict and update target location, which solves tracking target loss caused by serious shading problems.(3) A solution of multi-objective tracking has been designed. Kalman filtering shows the inefficiency in nonlinear and non-Gaussian system due to the complication of motion scene and motion estimation. Particle filter algorithm which is studied to realize target tracking algorithm based on particle filter of multi-feature in combination with the sample importance re-sampling method fusing weighted target color histogram and gradient direction histogram. CAMShift algorithm is introduced to optimize particle filter algorithm. It improves real-time and robustness of track.(4) A simple and efficient strategy for intelligent car control command is given to realize intelligent car to go along, stop, turn left and turn right by the feedback result of visual tracking algorithm. Meanwhile, a kind of control strategy of high-speed spherical camera in the direction and speed is designed. Under the situation of the sphere camera's mechanical parameter unknown, camera speed ratio is set manually, and camera speed is adjusted linearly according to the degree of deviation of target position. It can prevent the shocks of the spherical camera, so direction posture angle of P/T is adjusted through controlling it to rotate intermittently to maintain the target visible in the scene.(5) Under VS2008 development environment on the Windows Platform, using OpenCV image processing library, a set of hardware and software test system is established on moving object detection, identification and tracking, and control strategy. Many functional modules are integrated under one interface to accomplish intelligent car visual tracking system based on MFC library. And the effectiveness of various system functions is verified.