Vehicle Collision Warning System Based-on Machine Vision

In recent years, the number of cars in China is fast increasing, and the traffic accident rate are increasing as well. It is urgent task to reduce the number of road traffic accidents and to improve the safety level of road traffic for the current social. Current vehicle collision warning systems generally adopt laser, millimeter wave radar, ultrasonic and infrared hardware. Some of the world's leading auto company use mounted radar distance measurement system in top of the vehicle body. In the current warning system the hardware cost is not only high, but also complexity in processing of warning. The warning system is difficult to be embedded and integrated in products so that it is difficult for marketing promotion. Based on the analysis and comparison on the basis of various algorithms and implementation methods over the world, a kind of monocular vision algorithm is putted forward as the core of the on-board camera calibration technology. Science pavement in front of the car environment can be regard as a plane, the points on the pavement can be mapped on the points in the related image. After the analysis of traditional calibration, a general regression neural network is used to the on-board camera calibration because the network has good function approximation ability. Therefore, a relatively low cost camera can be used to achieve the front car distance measurement. Based on realization of front vehicle detection and tracking, and calibration algorithm, the distance of the front vehicle can be measured and collision warning system is designed. Experiments show that this system can greatly reduce the cost of existing ranging system, and can be embedded into the existing automotive vehicle traveling data recorder.This software is developed using virtual machine as the platform for the Ubuntu system in the PC. Qt and OpenCV2.4.9 programming are used to realize the algorithms of vehicle detection and tracking, and front vehicle distance measurement, and the all programs are translated to Linux system in machine vision gateway of XC-4412.