Research For Foot 3D Measuring System Using Multi-view Stereo Vision Based On Color Coding

In modern society,demand for shoes gradually transfer from basic functionality into a compound property,such as beauty,comfort,health and basic functionality.The traditional footwear industry can not meet people's growing needs.Customization technique can effectively compensate for these shortcomings.The existing customized shoe last CAD system is mostly based on laser scanning reconstruction.It is expensive and has a bottleneck in promotion.When talk about other 3D reconstruction techniques,structured light reconstruction has advantage of high maturity and low cost.However,the range of structured light reconstruction's applications is limited,because it is susceptible to ambient light.Multi-view stereo vision reconstruction is very complicated,but it costs less and collects images faster than other methods.It meets the requirements of the customized shoe last CAD system.Therefore,the use of multi-view stereo vision for foot reconstruction has received more and more attention.In order to obtain accurate parameters of foot,this paper propose a color-coded foot 3D measurement method using multi-view stereo vision.It provides more accurate interior and exterior camera parameters through sphere calibration.In combination with the idea of structured light reconstruction,a color rhombus array with good window characteristics is printed on socks to achieve a denser feature point matching.So our 3D measurement method can solve two problems.First,when checkerboard calibration applied to multi-camera,it will result in a larger cumulative error.Second,few significant feature points make it difficult to build dense foot models.By setting up a multi-camera platform with a uniform light environment and realizing synchronous acquisition of image,the quality of the data set is guaranteed.After analyzing the imaging process,the coordinate of the sphere's center is calculated by the circle detection.The 2D coordinates of the points on the calibration sphere are detected by the algorithm whice combines the filtering template with the gradient intensity.According to the geometric constraints of the sphere,the 3D coordinates of the points are calculated.After obtain the initial exterior camera parameters,SBA is used for nonlinear optimization to obtain the interior and exterior camera parameters.After studying the principle of pseudo-random array generation,the color diamond pattern is generated.Then we extract diamond corners and locate the decoding region.The HCL space is used for color recognition to decode these corners.Dense feature points are matched by epipolar constraint and code value constraints.The 3D coordinates of these corners are calculated by using the camera parameters and the matching relation.The Poisson surface reconstruction is performed to obtain a dense 3D model of the foot.By recording the stopwatch in running time,we can conclude that the synchronization of multi-camera is well and the time difference is within 10 ms.Using the above method to reconstruct foots,the integrity of each reconstructed feet is well.The errors of length and width are within 1 mm.At the meanwhile,the error of three circumferences and heights are within 2mm.