Vehicle Blind Spot Warning Algorithm Research And System Design Based On Monocular Vision

It can be found that vehicles blind area is an important cause of traffic accidents after analyzing the present traffic situation. Therefore, it is urgent to monitor the traffic of blind spot and give the driver early warning. Development of safe and reliable blind spot warning system has the very vital significance for the development of automotive engineering, traffic safety and security of lives and property.Firstly, after researching the traffic accidents and aiming at the place where the accidents has high frequency, the complete concept of blind spot warning system is proposed based on the traditional blind area. It includes city roads and the highway. Blind Spot Warning Algorithm based on Monocular Vision is proposed in this paper which has two parts:bicyclist detection in blind spot of vehicle rear mirror on the city roads and curve recognition algorithm on highway.This dissertation follows a two-step method to detect bicyclist:1. Hypothesis generation, where the location of possible bicyclist is hypothesized with the motion information.2. Hypothesis verification in which part tests are conducted to verify the existence of the bicyclist with the information of color, texture and so on. The aggregated channel features are defined as the bicyclist representing features and machine learning method is used to get the classifier which is used as the criterion in online detection.In order to recognize the curve of highway, lane is modeled at first. Taking both accuracy and quickness into consideration, the straight line-curve model is adopted in this paper. Then Hough transformation is used to detect straight part of the road and straight-line deviation method is used to judge the direction of the curve.In the end, DaVinci embedded system based on TMS320DM6467 processors is used to perform the blind spot warning system. The main work of this part includes the hardware design, software design and algorithm transplant.