In order to solve the problem of mutual influence among the estimations of each coordinate caused by the canonical transform used for decoupling high-dimensional coupled kinematic state models in maneuvering targets tracking applications, an improved decoupling method is presented. At first, explicit expressions for unbiased compensation coefficients and unbiased covariance statistics based on Kalman filter predictions related to the 3D measurements are given. And then, based on the canonical transform, an improved decoupling method using the modified weighted matrix is presented in detail. At last, simulation experiments are conducted combining with the IMM (interacting multiple model) algorithm. Results indicate that the proposed algorithm can reduce computational burden and eliminate influences among three Cartesian coordinates, which is good for analysis and application of IMM algorithm.