Research On Key Technology In High Accuracy Binocular Ranging

Binocular vision has been widely in recent years.Binocular vision is applied in manufacturing and human life,and has important applications in aerospace,transportation,robotics and other fields.binocular vision imitates human eyes in order to achieve 3D reconstruction of the scene.The classical binocular stereo vision includes four processes,namely stereo calibration,stereo correction,stereo matching,and 3D reconstruction.The focus of this paper is to study the key technology that affect the accuracy of ranging and to propose relevant method to reduce the error.Therefore,the analysis is made in the following three directions.The structural parameters in the binocular vision system have an important influence on system performance and ranging accuracy.We will analyze the design requirements of binocular vision high-precision ranging system.Firstly,we will study the imaging model of binocular vision.Secondly,we will research on the influence of structural parameters on the visibility and horizontal field of view of binocular stereo vision system.We will also study the influence of structural parameters on ranging error.Lastly,we will get some guideline about designing high-accuracy ranging system.We will research on Zhang's calibration method based on radial and tangential distortion for stereo calibration process.In binocular vision,the purpose of stereo calibration is to calculate the structural parameters of the binocular stereo vision system.If the parameter structure error calculated by the stereo calibration algorithm is large,it will inevitably affect the accuracy of the ranging.Firstly,we analyze the principle of Zhang's calibration method.Considering the distortion model in Zhang's calibration method is simple,we propose Zhang's calibration method based on radial and tangential distortion model,and the effect of the improved calibration algorithm is verified by experiments.We will research stereo matching algorithm based on deep residual network.In binocular vision,stereo matching has the greatest impact on ranging accuracy.This is because once the matching algorithm is mismatched,the depth error will be proportional to the square of the distance,which is much larger than the ranging error caused by the structural parameter error.We apply the deep residual network into stereo matching so that we can increase the number of layers without the situation where the gradient disappears,improving the matching accuracy.