Vehicle collision warning system can determine the relative distance and speed between target vehicle and the front vehicle by monocular vision positioning technique from automobile license plate image captured by camera so as to judge danger level and remind the driver to make appropriate action and avoid vehicle collision timely. Study on the positioning technology of this system aims to help the driver to judge and improve driving safety. Thus, the system has a broad application prospect. The research content of this paper could enrich and supply PNL visual locating method, endowing with significance of theoretical research. The paper proposes an improved vehicle measuring method based on monocular vision for vehicle license plate. This method combines the characteristics of fast speed for analytical solution method and high positioning accuracy for iterative solution method, therefore has a high robustness and overcomes the multi-solution problem of P3P iterative method. The simulation experiments show that localization precision of the improved positioning method has been improved greatly as compared with P4L method. At the same time, the real-time characteristic of collision avoidance warning system with improved visual locating method has been improved a lot, and the new location algorithm has good performance in real-time characteristic, which greatly improve the processing ability of the system for images.