Research Of Sustained Tracking Targets During Immobile And Moving Status Methods Based On Multiple Statistical Methods

The main goal of this paper is to study the continuously track, under the complex trend, of dynamic and static multiple targets. The moving objects in visual sense are our primary research subjects.First, the proposed video frame preprocessing algorithm, which is based on "noise inversion", can both remove random inter-frame noises effectively and preserve the continuous moving inter-frame targets.Then, on the basis of conditional statistical background subtraction, the paper put forward the improved motion recognition algorithm based on images of movement history. This algorithm gets the target foreground by conditional statistical method. The binarizational picture is used as the input picture. After segmenting the motion histroy image, based on the spatio-temporal relativity, the algorithm deal desnoising mathod with the individual athletic modules. By calculateing the motorial direction whit each module, the algorithm get the histogram. The integreted motorial condition of the target can be analysised and estimated with histogram. In the experiment, this algorithm is proved efficient for motorial tarcket tracing. And the imprved algorithm is better than the original one in temporal complicacy and real-time.The third, the modeling algorithm of self-adaptive statistics is proposed, which is on the basis of improved Camshift algorithm, for static targets. This algorithm gets the retion of interested by miproving and weighting the Camshift algorithm. This retion is used to model the adaptiv features. With the records of model, altorithm can much the next model when the same target stop again. The validaty and adavance of this algorithm with the static target is also prooved in the experiment.At last, in order to solve the transform state caused by the same target during in moving state and static state, the paper put forward the tracking association function which can effectively and continuous track multiple targets and the performance is superior to other same algorithms attested by experiment.