Accurate and real-time acquisition of vehicle speed information is a necessary prerequisite for vehicle to achieve high-precision positioning and navigation, advanced cruise control and formation cruise control. Therefore, a novel vehicle speed estimation method was proposed based on UniTire model. Firstly, a vehicle speed estimation nominal model, which includes vehicle longitudinal, lateral and yaw dynamics, was established based on UniTire model. The global accurate solution of characteristic parameters of the model was obtained based on adaptive genetic algorithm and quasi-Newton algorithm. Subsequently, the process noise and observed noise of the nominal model was taken as scheduling variables, and a cluster of vehicle speed estimators was designed based on strong tracking central difference Kalman filter. The outputs of the cluster of vehicle speed estimators were smoothly fused based on interactive multiple model algorithm. Finally, the feasibility and validity of the proposed self-driving vehicle speed estimation method were verified with hardware in the loop simulation system for full-vehicle. The results show that, compared with strong tracking central difference Kalman filter, the proposed method has stronger adaptive ability to the statistical characteristics of the system noise and higher estimation accuracy.