Abstract: It is rather difficult to measure the longitudinal velocity of four-wheel drive vehicle directly. Considering the dynamic confidence feature of signals from different sensors, an adaptive longitudinal velocity estimation method based on kinematic information fusion is proposed. The driving environment influence on vehicular sensors is investigated. Then the kinematic model for data fusion is established under Kalman filter framework. The adaptive algorithm regarding with the colored noise is designed to compensate the non-random and time-varying disturbance oriented from the slip ratio and road grade. To make better balance between the stability and the optimization, the strong tracking filter with fading factor is introduced to suppress the divergence under critical conditions. By Carsim/Simulink platform, the proposed method is validated and compared with the baseline method, such as H_∞ filter, under hill acceleration, wheel slip and double lane change conditions. The processor-in-the-loop system is developed to verify its execution time and consistence. Results reveal that the proposed method can accurately and stably estimate the longitudinal velocity with good real-time performance. Compared with the baseline methods, it shows better adaptability and has no requirement for prior statistics knowledge of noise. Thus, the high accurate velocity information for four-wheel drive vehicles can be obtained by this unified method under complex conditions, which is especially significant for vehicle active safety.

Key words: four-wheel-drive vehicle, longitudinal velocity estimation, Kalman filter, noise adaptability, kinematic information fusion