Adaptive estimation method for pavement adhesion coefficient integrating visual and dynamic data

The road adhesion coefficient is widely recognized as a key factor affecting vehicle active safety control. Accurate estimation of this parameter faces challenges such as visual interference and weak observability of vehicle dynamics under low excitation conditions. This paper presents an adaptive estimation method fusing on-board vision and vehicle dynamics. First, a seven-degree-of-freedom vehicle model and the Dugoff tire model are constructed, and an Unscented Kalman Filter (UKF) observer is designed for dynamic inversion. Second, a lightweight road recognition model is established based on MobileNetV3 to provide prior information on road surface type. Finally, a Sigmoid-based adaptive weighting strategy according to dynamic excitation level is proposed to realize dynamic fusion of vision-based and dynamics-based estimation. CarSim-Simulink co-simulation results demonstrate that the proposed method achieves smooth switching under varying excitation conditions and maintains stable estimation performance under the simulated vision-failure condition. Specifically, the visual model achieved an average classification accuracy of 96.69%, while the fusion algorithm reduced the root mean square error (RMSE) by approximately 65% under low-excitation conditions and by approximately 51% under varying road surfaces and vehicle speeds compared with the pure UKF method. In addition, the proposed strategy outperforms a confidence-based adaptive fusion benchmark in excitation-deficient and vision-degraded scenarios. This study demonstrates the effectiveness of the proposed fusion framework in simulation-based validation and provides a promising solution for road adhesion coefficient estimation under varying driving conditions.