KLT Target Detection And Tracking In Fractal And GVF Snake Based On

In order to provide relevant technical support for the visible light TV seeker, this paper investigated automatic detection, recognition and tracking algorithms of moving target. The main contents are as follows:Firstly, an algorithm was proposed based on fractal theory and improved GVF Snake model for artificial target image detection. The initial contour of the Snake model was identified through fast differential box-counting algorithm. The real contour of the target was determined on the basis of improved GVF model. Simulation showed that the proposed algorithm overcame the shortcoming that general fractal algorithm scans by points at low efficiency. The contour of the target image was better extracted from natural background.Secondly, the paper discussed feature extraction methods of the target image. On the basis of analyzing the artificial target, the feature is selected by contour feature, affine invariant moments and Hu invariant moments feature. Secondly, investigating the classifier algorithm principle, classification, characteristics and the selection of kernel function for support vector machine. Finally, the artificial targets of aircraft and tanks were recognized by support vector machine classifier. Simulation showed that the classifier above was of higher recognition accuracy.Thirdly, we focused on object tracking which was based on KLT algorithm.To improve the accuracy of tracking, the pyramid algorithm was successfully applied to tracking algorithm of feature points. For the problem that feature points would lost in the process of tracking, the extraction algorithm of feature points was improved. Finally, the improved KLT tracking algorithm was transplanted into TMSDM642processors and a series of optimization was carried out from hardware and software of DSP. Experiment showed that the migration algorithm is reliable.In this paper, the detection and tracking algorithm of moving target was investigated. And the tracking algorithm has been successfully transplanted to DSP, realizing the stable and reliable moving target detection and tracking test.