Research And Design Of Motion Targets Detection And Tracking System For Mobile Robots In Unknown Environments

Moving targets detection and tracking in video sequences is one of the hotspots that having been widely researched in computer vision and image processing fields. This thesis focuses on the detection of single object motion in complex environments and intends to present a flexible and adaptable motion target tracking system that can be applied to mobile robotics fields under unknown environments.Kinematics of the camera is first analyzed in this thesis. Then a space plane Z_c =1 with unit distance off the camera optics center is presented for mapping all the space points to it, so that real distance between the camera and the target can be omitted thus predigests the kinematics equation.Due to the less efficient differencing or background subtraction method singly in variable environments, a combination of three-frame differencing and adaptive background subtraction technique with motion compensation is adopted to detect moving targets. And the priori estimate of Kalman Filter is used to compensate the motion. Experiments prove that this method can better compensate for the backgrounds motion during the tracking process.For more precisely tracking and more smoothly control, an Extended Kalman Filter (EKF) suitable for robot system is constructed in this thesis on the basis of traditional Kalman Filter, which estimates the object's position in the next time of day, meanwhile makes the correction by use of the observation data. Accordingly an estimation-correction-controlling closed loop is implemented for tracking.In the process of seeking moving center, prediction data of EKF is used to determine the first clustering center to cluster the difference results thus efficiently reduces the iterative process. Because of the great computing complexity and accumulative error of the average targets template updating strategy, a difference template updating strategy is presented in this thesis to judge the tracking status that preferably solve the above problems.Experiments prove that the scheme designed in the thesis have preferable detection and tracking effects on single object moving with biggish background shift between frames, which display validity and feasibility of the above system.