In order to reduce the influence of random errors and outliers on the accuracy of vehicle laser Doppler velocimeter, an adaptive Kalman filter algorithm is proposed. Based on the "current" statistical model (CSM) and combined with the actual characteristics of vehicle velocimeter, the state-space model of system is built, and the adaptive adjustment of acceleration variance is realized by the deviation between measured and predicted value of speed. The adaptive algorithm for measuring noise variance, which can eliminate outliers and reflect the real-time characteristics of road, is given according to the orthogonal properties of innovation and speed estimation error in Kalman filter algorithm. Simulated results show that the algorithm is better than CSM algorithm in the convergence speed of filtering and estimation accuracy. Experimental results show that this algorithm can significantly improve the accuracy and robustness of velocimeter.