Multi-target high-precision continuous tracking method based on state regression Transform architecture

The invention discloses a multi-target high-precision continuous tracking method based on a state regression Transform architecture, and the method comprises the steps: firstly, designing and obtaining radar simulation data, generating two-dimensional measurement, carrying out scene preprocessing, extracting features from X-Y coordinate measurement, inputting the features into a high-dimensional vector, inputting the high-dimensional vector into a constructed SR-MT3 network model, carrying out the training of the network model, and carrying out the calculation of the high-dimensional vector. And finally, inputting test data into the trained network model to obtain a multi-target estimation tracking result. According to the method, from the perspective of deep learning data driving, the limitations of model dependence, data association difficulty and the like of a traditional Bayesian method are overcome; the problem that continuous tracking cannot be carried out by an original MT3 algorithm, the problem of sl