Research And Design Of Driver Assistance Algorithms Based On Stereo Vision

The growth and development of road transportation is faster than ever before.Obviously,safety is one of the most crucial aspects of road transportation.Advanced driver assistance system,or ADAS,has become an important part of intelligent transportation system while the researches on this field become increasingly active as well.Computer vision based ADAS,due to its advantages of high integration and low cost,provides an important research direction of ITS and ADAS.The driver assistance system based on stereo vision is studied thoroughly.On the basis of evaluating the development of computer vision based ADAS in China and abroad,vehicle detection algorithms,stereo vision matching algorithms and anti-collision warning algorithms are explored and designed.1.Feature extraction method based on image gradient is studied.The method preserves almost all the vehicle feature information and effectively reduces the feature dimension using histogram of oriented gradients.Combining the image gradient feature and linear SVM classifier,image pyramid algorithm is implemented to achieve the multi-scale vehicle detection in real-time.2.The principle and process of stereo vision is explored,and a stereo vision algorithm based on deep learning is designed.Using deep learning,a convolution neural network is trained to calculate the stereo matching cost.Combining CNN and traditional algorithms like semi-global matching and bilateral filtering,the result of stereo depth map achieves high accuracy.3.The statistics of vehicle acceleration during driving are studied and summarized,and an anti-collision warning strategy is designed.Combining real-time vehicle detection and stereo vision,an anti-collision algorithm based on vehicle velocity is implemented.Ultimately,the complete design of driver assistance system is analyzed and verified.As demonstrated by the test results,the driver assistance system based on stereo vision can perform vehicle detection and anti-collision warning accurately,the system is stable and achieves real-time performance.