This paper researches an autonomous driving system for exploring, planning, and tracking reference paths in unknown environment. Firstly, the classic computer vision algorithm processes the discrete obstacles in the unknown environment and passes the valuable information into the neural network to control the vehicle driving. Then, sensors such as Lidar and IMU are fused to construct a grid-based map and plan the global trajectory. Finally, the MPC based on three-degree-of-freedom vehicle motion is established to track the global trajectory, while Simulink/CarSim simulation and debugging of the control process are carried out, and the autonomous vehicle is built to test in the natural outdoor environment. The results show that the autonomous vehicle explores the information accurately and reliably during the simulation and test while the trajectory tracking is stable.