Pavement Disparity Robustness Detection And Shadow Disparity Correction Method

Binocular stereo vision is one of the core technologies of intelligent driving systems.Obtaining high-precision disparity maps from road scenes by binocular stereo matching is very important for depth sensing,3D reconstruction,etc.,and has been the focus of researchers at home and abroad in recent years.With the gradual improvement of industrial requirements,the requirements for disparity accuracy are also getting higher and higher,which makes the error analysis of binocular stereo vision system especially important.In the complex urban traffic driving environment,failure to accurately acquire binocular disparity will affect the judgment and operation of intelligent driver assistance systems on road scenes,and may cause traffic accidents in serious cases.In order to clarify the influence of various factors on the road disparity and improve the accuracy of binocular stereo vision,this paper carries out the research on the robustness detection of road disparity and shadow disparity correction.The main research contents and innovations of this paper are as follows:1)The influence of the calibration accuracy of each parameter of the binocular system on the stereo calibration results was analyzed by adding a small perturbation 8 to the internal and external parameters of the binocular camera,and studying the trend of the derivative function of the coordinates of each pixel point of the left and right images with respect to 8 by the control variable method.2)A new quality evaluation index for pavement disparity maps,i.e.,the standard deviation of pavement V-disparity fitted straight lines,is proposed.This evaluation index does not require a standard disparity map as a reference,and is general in nature.In this paper,the effects of illumination,noise and internal and external camera parameters on pavement disparity are explored.3)A shadow disparity confidence calculation method is proposed.Firstly,the shadow boundary is segmented from the pavement image,and then the disparity is recalculated with a larger sliding window on the shadow boundary,and the calculated disparity is compared with the original disparity,and the original disparity is assigned a confidence level according to the comparison result to achieve an accurate prediction of the pavement scene.4)According to the confidence magnitude of the disparity at the shadows of the road,the disparity with lower confidence is removed,and the missing values are interpolated and filled using the disparity of the non-shadowed road at the same distance,and after processing,the height at the shadows is significantly reduced,which effectively reduces the interference of the shadows with intelligent driving.To verify the effectiveness of the method in this paper,a series of simulation and real experiments are carried out under different driving scenarios.The experiments show that the method of this paper comprehensively and systematically analyzes the influence of calibration accuracy of internal and external parameters of binocular system on stereo correction results,as well as the influence of each factor on the accuracy of road disparity,effectively reduces the interference of shadows on the automatic driving system,and can play a guiding role in the design and use of in-vehicle binocular vision system.