Abstract: Accurate and fast vehicle distance measurement is the indispensable data basic for vehicles to achieve autonomous driving and obstacle avoidance control, and it is also a necessary component for safe driving of intelligent vehicles. Based on the camera imaging principle and the Yolo only look once(YOLO) deep neural network, a fast and effective monocular camera ranging method is proposed, i.e., four-point calibration distance measurement algorithm. This algorithm combines the deep neural network and the ranging algorithm to quickly identify the target vehicle and calculate with the self-vehicle. In terms of the vehicle identification, the detection box is collected through the Common objects in context(COCO) data set and the YOLO network model. To obtain the information of the relative distance, the dynamic vertical and horizontal camera ranging model are established according to the principle of camera pinhole imaging. Meanwhile, the analytic equations of pixel coordinate system in YOLO network detection box and the world coordinate system in actual ranging are established, and the fast four-point calibration distance measurement method is proposed. Finally, the proposed four-point calibration distance measurement method is compared with the internal parameter calibration ranging method in the experiments. The experimental results show that the measurement error is reduced by about 1% within the range of 25 m, and the complex internal parameter calibration steps can be omitted.

Key words: vehicle distance measurement, neural networks, autonomous driving