Research On Real-time Moving Object Tracking Algorithm Based On Machine Vision

With the rapid development of computer technology and digital image processingtechnology, motion tracking of video sequences is currently a hot research, and it includingpattern recognition, computer vision, artificial intelligence technology, has been widelyapplied in various fields. At present, much attention has been paid to moving target trackingalgorithm based on the theory of the filter, and has been widely used. Kalman filter is oftenused to predict the position of moving objects and other features of the data, and the algorithmcan obtain good tracking effect in linear system. But the error of Kalman Filter will becomelarge or even diverging when the target status changes suddenly. For KF Filter algorithm isonly applicable to linear systems, There has been extended Kalman filter algorithm, But EKFonly suitable for weak nonlinear system, and the performance of EKF is not stable, or evenfiltering divergence in strongly nonlinear system.In view of the above requirements, in this paper, when the moving target state mutated,the effect is not ideal based on KF and EKF filtering algorithm. So, the further research andimprovement has been done. The concrete research content is as follows:（1）Studied the Kalman Filter algorithm and multi-innovation identification theory indepth, proposes a new improved Kalman filter algorithm based on multi-innovation theory(MI-KF). MI-KF has better precision and stability, because MI-KF takes not only the movingtargets’ current state of motion into consideration, but also the time before, Further, theconvergence properties of the proposed Multi-innovation Kalman Filter has been proved intheory. Finally, simulation studies were carried out respectively from two aspects of curvesimulation and video sequence, and simulation results show that the improved algorithmMulti-innovation Kalman Filter is superior to the traditional Kalman Filter.（2）Combined with multi-innovation theory, proposed the MI-EKF. On the basis of thestandard EKF，MI-EKF has more innovations, and it also considering the multi-stepmovement in the process of target motion information, which has greatly improved filteringaccuracy of the algorithm. In the simulation of algorithm, we conducted experiments MI-EKFinclude two innovations and MI-EKF include three innovations respectively, and the results are analyzed.（3）Through the actual project, build a remote servo control system based on machinevision software and hardware platform. Combined with hardware, such as PLC, servo motor,and WebAccess realized the remote control of servo motor. In the system, the movementtarget tracking using MI-KF algorithm proposed in this paper. Through practical application,the improved algorithm proposed in this paper can meet the requirements of engineeringapplication.