Adaptive tracking method of biomimetic-pattern recognized targets

The present invention discloses a biomimetic pattern recognition adaptive target tracking method, mainly to solve the prior art is poor robustness, high diversity training samples requested problem. Its implementation steps are: (1) obtaining training samples; (2) training samples extracted characteristics; (3) to establish the basis of training gray feature super sausage neural network, based on the training sample number and characteristics of the training sample size characteristics established Radial Basis neural network; (4) to calculate the Euclidean distance super sausage neural network and the search area to all candidates between areas; (5) to obtain the target position with a minimum distance method; (6) to obtain the target position in the ultra-sausage neural network ID, and inputs the RBF neural network to obtain the target size. The invention of the training set of training samples overlay training to improve the tracking robustness, reducing the dependence on diversity training samples, can be used in intelligent robots, intelligent transportation systems and video surveillance, and other fields.