In the context of unmanned multi-vehicle formation guided by manned vehicles, a system for vehicle recognition and trajectory tracking control of unmanned vehicles during formation driving was devised and executed. An algorithm for multi-sensor fusion moving target detection was proposed, leveraging data from lidar, camera, and mmWave radar sensors. The algorithm utilizes Euclidean clustering, deep learning, and kinematic reasoning techniques for target detection. Additionally, a fusion methodology was introduced to integrate detection outcomes from various sources for precise identification of vehicles in the vicinity. Paths were anticipated based on the trajectories of preceding vehicles, and a Kalman filter was developed to smooth and filter these paths. A vehicle dynamic model, vehicle road error model, and the robust H∞ controller was established for vehicle trajectory tracking control simulation. Outcomes from simulation and real vehicle validation show as follows. The average recognition accuracy of preceding vehicles in test scenarios exceeds 95%. The mean squared error and average trajectory deviation rate of real-time anticipated paths decrease by 17.3% and 48.6% respectively pre and post filtering. Lateral control position error and yaw angle error decrease by 29% and 41% correspondingly compared to PID control. Vehicle formations attain stable working at speeds of up to 54 km/h.