Estimation of Object Motion State Based on Adaptive Decorrelation Kalman Filtering

To estimate the motion state of object feature point in image space, an adaptive decorrelation Kalman filtering model is proposed in this paper. The model is based on the Kalman filtering method. A first-order Markov sequence model is used to describe the colored measurement noise. To eliminate the colored noise, the measurement equation is reconstructed and then a cross-correlation between the process noise and the newly measurement noise is established. To eliminate the noise cross-correlation, a reconstructed process equation is proposed. According to the new process and measurement equations, and the noise mathematical characteristics of the standard Kalman filtering method, the parameters involved in the new process equation can be acquired. Then the noise cross-correlation can be successfully eliminated, and a decorrelation Kalman filtering model can be obtained. At the same time, for obtaining a more accurate measurement noise variance, an adaptive recursive algorithm is proposed to update the measurement noise variance based on the correlation method. It overcomes the limitations of traditional correlation methods used for noise variance estimation, thus, a relatively accurate Kalman filtering model can be obtained. The simulation shows that the proposed method improves the estimation accuracy of the motion state of object feature point.