To deal with the problem that the precision of weighted data fusion may decrease when measurement noise varies

a novel algorithm of variance estimation is proposed. Variance estimation is implemented by adaptive moving data window

length of window is determined by hypothesis testing result. In the part of hypothesis testing

statistics of the hypothesis testing approximated the normal distribution by utilizing signal segmenting processing method and the central limit theorem

which simplifies the following calculation and theoretical deduction. Then

based on the Markov state transition theory and the maximum a posteriori criterion

the change of the measurement noise variance is detected quickly. The comparison of simulation results verifies the improved algorithm overcome the limitation of the typical method and guarantee higher fusion precision.