Research On Matching Technology Of Machine Vision Danger Warning System With Passenger Car

Considering "zero accident, zero death" as the ultimate goal of vehicle safety, driving assistant system is a main technical technology. Lane departure warning system and forward collision warning system are an important part of a driving assistant system. The correct rate and false warning rate of the system are not only an important characteristic to evaluate the system`s performance, but also an important factor affecting subjective driving experience of the system. On the other hand, research on integration of other driving assistant functionalities is a development direction of the system.Facing the increasing severe market situation, it`s necessary to develop the vehicle level matching capability of driving assistant system and to improve Chinese automobile independent brands` car safety and competitiveness. Basing on the integration of machine vision technology, this research is focused on main steps of vehicle level matching, such as performance test、performance evaluation、sensor calibration etc.For danger warning system basing on one-camera machine vision technology, a test platform has been designed and constructed. The video image data and vehicle`s operating mode information have been collected as test samples; the expectation has been extracted from test samples as evaluation criteria using upper computer software; and then a Hardware in Loop test platform has been constructed. The test results and analysis report have been established by importing test samples into the driving assistant system and by comparing the recording results with the expectation. This test platform could also be used to compare different driving assistant systems for distinguishing problematic operating modes, and establishing lane departure warning system and forward collision warning system`s evaluation standard.The research has introduced the black and white external reference calibration method of camera. The least square method for approximate solutions and Weinberg nonlinear optimization method for the minimum values is used to simplify the calibration model, to improve the calibration efficiency and to reduce interference of calibration environment on corner detection. This calibration method is used to meet the requirement of off-line fast and reliable calibration. The reliability of the calibration method has been approved.Finally, the work studied the control system of automatic headlights basing on machine vision image. The initial research work has been done on sensing technology of automatic headlight control using image recognition: including the extraction of the lighting blocs using image segmentation method;the calculation result of lighting blocs using the Kalman filtering method;recognizing the taillignts of forward vehicle and the headlights of vehicle coming in the opposite direction,and also street lights.