Indoor positioning and tracking, which can make the available space be effectively allocated and navigate the police, soldier, doctor, etc. to fulfill the special indoor task, has attracted considerable attention.Firstly, the advances of the current indoor positioning system are reviewed and the existing problem pointed out. To mitigate the side effect of NLOS and multipath transmission in the mobile localization, node localization is embedded into the framework of machine learning. Node localization is carried out by extracting radio frequency characteristics and using kernel function to build the relation between radio frequency characteristics and node positions. TOA and RSSI are only as characteristics in the method, not as accurate distance estimation, so the method can extremely reduce the bad effect of NLOS transmission on localization error.Due to support vector regression corresponding to the Chebyshev center of feasible region, it leads to reduce the its generalization performance when feasible region is asymmetry or strip. Then we analyzes the lackage of current regression algorithm, and presents a Non-linear regression algorithm based on feasible region analytical center. We Theoretically analyzes the relation between the regression algorithm and maximum-likelihood parameter estimates, gives its iterative steps, and uses the algorithm to localization for indoor nodes.In the application of mobile track, noise doesn't satisfy Gaussian noise model, so the mobile track of kalman filter algorithm has some difficulties to get the precision required. Particle filter has little restrictions to noise, but the computation of particle filter algorithm is very large, and particle degradation occurs in the course of iteration. In the practical application, it is very difficult to achieve noise statistics or noise modeling. So the paper presents a mobile position track algorithm based on game theory. The algorithm sees noise modeling as game opponent, and the noise serial game opponent generated contains random noise and estimate error of certainty. By differential game theory, it is mobile track according to saddle point of the game objective function. The simulation results show, compared to kalman filter and particle filter algorithm, the performance of mobile position track algorithm based on game theory has extremely improved. |