Person tracking problem under the indoor environment often face light, occlusion and similar color interference, in order to enhance the tracking stability and accuracy in the situation, this paper proposes a root mean square (RMS) unscented H-infinity filtering method in Considering interference noise perturbation. First of all, Unscented transformation is used to replace the complex Jacobi matrix calculation under the framework of extended H-infinity filter. The method, which using Gaussian density approximating filter distribution, reduce the linear approximation error, and reduce the interference of system model noise to the estimation values. Second, the Cauchy decomposition to the state covariance matrix RMS, which uses diagonal elements to calculate, reduce the disturbance of observation noise error and filter calculation consumption. Video tracking experiments showed that compared with the EKF、UKF and PF algorithm, the proposed algorithm can effectively improve the pedestrian tracking accuracy in mutation, and reduce the time consumption.
