Research Of Moving Target Tracking System Based On Visual Panorama

Moving target detection and tracking technology are always one of the hot contents in computer sight field, and also core algorithm of intelligent video monitor system. Basing on kernel density estimation of non-parameter model the moving target detection has performances such as good results, high flexibility and good suitability. Besides, with the advantages of no parameters initialization and complex parameters updating algorithm it has a big potential in applying real time target tracking field. The research content in this article is target detection algorithm, and one target detection algorithm of good performance is put forward, finally it is verified through test.First the main moving target detection algorithm which is mixed difference time domain method, Gaussian mixture model approach and kernel density estimation detection algorithm is deeply analyzed in the article. Then the performance test is carried out by experiment, and reliable data are obtained which are used to make contrast analysis to every algorithm. The results are advantage, disadvantages and applying occasions which are used to provide real basis and experimental bases for improving algorithm.Secondly to overcome the disadvantages such as large amount of computation and low real-time of kernel density estimation detection algorithm, one moving target detection method basing on typical sampling and diversity weights is given. This method abandons initial sample and decrease computation of algorithm. While to make up negative influence of sample lost, diversity weights algorithm is put forward to provide supplement with this algorithm. Aiming at special situation of changing monitoring background, on new sample updating method is put forward which not only can update sample information but also enhance anti-jamming ability according to changing background. According to contract between kernel density estimation algorithm and sample, one threshold segmentation algorithm basing on sample is put forward which can merge with kernel density estimation algorithm well. Finally, effectiveness of this improved is verified through test.The imaging principle of panoramic camera and dynamic performance of PTZ camera are further analyzed. The panoramic and PTZ coordinate system are established, and the change between two coordinates are realized. PTZ automatic focusing,rotary speed control and algorithm for prediction of target trajectory are realized. The PTZ tracking lag deviation of PTZ turning lag is corrected to improve accuracy of PTZ tracking effect.Finally, under the condition of every targets' moving orbit, through test verifying actual effect of PTZ tracking and then contrast analysis of data, the effectiveness of algorithm is proved. And under some special condition, the weakness of algorithm is pointed out and some improvement is given for guidance of further research.