Research On Highly-accurate Structured Light 3D Reconstruction Technique Based On Photometric Analysis

3D reconstruction techniques and systems are the key components of machine vision system.Highly-accurate and high-resolution 3D reconstruction techniques contribute to the development of machine vision and artificial intelligence.Binarystripe-based structured light technique is one of robust and accurate 3D reconstruction methods,which can acquire micron-level reconstruction accuracy for Lambertian surface without texture.Whereas only degraded reconstruction results can be acquired for surface with non-Lambertian reflectance properties and texture.Given the effects of non-Lambertian reflectance properties,surface texture and complex geometry of surface on the accuracy of binary-stripe-based structured light technique,we proposed photometric-analysis-based structured light methods.With these regards,this dissertation provides the following contributions that address the above concerns.Firstly,on close-range photometric stereo,we propose multiple-sphere-based and reference-plane-based calibration methods of point light source to feature LED point light source,including its position,principle optical axis and main intensity.To feauture the illumination variances,we introduce the lighting model which considers both angle attenuation and distance attenuation.Based on the calibration results and the lighting model above,we acquire accurate the normal and albedo of surface,which is used to improve the reconstruction results of binary-stripe-based structured light technique further.Secondly,given the failure of stripe detection and noisy points in binary-stripebased structured light system due to the geometry of surface and system niose,a binarytree-based and scene-adaptive stripe detection method is proposed.By traversing the binary tree,the specific stripe is detected within the minimum searching margin instead of the whole row.The system noise and geometry of the surface outside the minimum searching margin have no effect on stripe detection,which improves the robustness of the decoding algorithm.In addition,without projecting the Gray code patterns,we propose photometric-stereo-based method to solve the ambigulity of each subregion.The global phase value of each subregion can be acquired based on the initial depth from photometric stereo system instead of Gray code coding patterns.Thirdly,based on the analysis on the reconstruction errors of binary-stripe-based structured light system due to surface texture and non-Lambertian reflectance properties,we proposed the fusion system consisting of close-range photometric stereo and binary-stripe-based structured light.To decrease the effect of non-Lambertian reflectance properties on the contour of stripe edge,for the first time,we introduce the albedo of surface to the detection of stripe edge with subpixel accuracy.By eliminating the non-Lambertian reflection components of the stripe contour,an improved detection result of stripe edge with subpixel accuracy is acquired firstly.Then the fusion algorithm of the normal vector from photometric stereo system and the depth information for binary-stripe-based structured light system is proposed to eliminate the effect of surface texture on reconstruction results and acquire camera-level reconstruction resolution.The content of this dissertation ranges from close-range phometric stereo and binary-stripe-based structured light technique to the fusion methods.The concerns of binary-stripe-based structured light technique are solved by introducing photometric analysis in the decoding processing of binary-stripe-based structured light technique.The above ideas contributes to solve the ongoing issue on the highly-accurate and highresolution reconstruction of complex scenes academically.In addition,three systems,including close-range photometric stereo,binary-stripe-based structured light and the fusion system,were completed as well,which have been used in practical scenes.