| Target detecting and tracking has been a very important research topic in the field ofcomputer vision, the Mean Shift algorithm is an excellent target detecting and trackingalgorithm. For solving the situation that most of the target is covered, this paper mainlystudies the target detecting and tracking algorithm which combined with the Mean Shiftalgorithm and the Kalman filter, then transplanted this algorithm to the OMAP3530hardwareplatform and completed an independent target detecting and tracking system.Mean Shift algorithm using nuclear color histogram as the model for describing target,the single peak of nuclear function allows the algorithm has better robustness to controlpartial occlusion or target deformation of the target, and is real-time. However, the Mean Shiftalgorithm has some drawbacks, such as target motion too fast or the target encounteredobscured by the large proportion, in these cases, the algorithm will converge to a backgroundregion which similar to the target color, resulting in tracking failure. In this paper, the MeanShift algorithm and the Kalman filter is combined and describing the specific methods.Kalman filter theory is briefly introduced, then specific discussion how to model the Kalmanfilter to make it has a good prediction function used in target tracking and improve thetracking performance of the Mean Shift algorithm for fast moving targets.Firstly, this paper introduces the OMAP3530hardware platform. Secondly, the paperdetailed introducing the construction of the Linux system and algorithm transplant, throughthe analysis of experimental results, the target detecting and tracking algorithm whichcombined with the Mean Shift algorithm and the Kalman filter has a stable running effect andhas a high efficiency. Through different conditions of video experiment and analysis ofexperimental results, the operation efficiency can be increased by15%compare with thesingle Mean Shift algorithm. At last, the algorithm was transplanted to the OMAP3530platform. |
PDF Full Text Request