Study Of Target Tracking Algorithms Based On Image Matching

Target tracking algorithm based on image matching is widely used in practicaltarget tracking systems because of its excellent performance. The matching trackingalgorithm is still effective under occlusion and distortion. On the basis of study andsimulation of the traditional matching tracking algorithm based on gray correlation, twotarget tracking algorithms based on image matching are centred on in this thesis. One isbased on normalized mutual information. The other is based on high-dimensionalbiomimetic neural network.Target tracking algorithm based on normalized mutual information has goodstability when the target is rotated, occluded and the illumination condition is changed.A new adaptive template update strategy is introduced, which is based on information ofmulti-scale image in scale space. In the improved algorithm, the template can beupdated adaptively with target size changing. Experimental results demonstrate thatstability and adaptability of this algorithm has been greatly improved.To overcome the intrinsic shortcomings of target recognition and trackingalgorithm based traditional artificial neural network, the tracking algorithm based onhigh-dimensional biomimetic neural network is proposed, which combines principalcomponent analysis (PCA) technique with the high-dimensional biomimetic neurons.And a matching function method and a template update strategy are designed, and theymake the algorithm more robust. Simulation results show that this algorithm also hasgood performance in noise, low contrast radio, rotation and occluded environment.