Research On Fringe Stereo Matching And Generation Algorithm In 3D Measurement

With the characteristics of simpleness,high accuracy and speediness,structured light three-dimensional measurement technology is widely used in many fields such as industrial inspection and machine navigation,and therefore has become a research hotspot in recent years.In order to meet the needs of developing compact,low-cost measurement system,this paper selects the domestic MEMS chip for projecting structured light.Compared with DLP projector,this MEMS chip has the advantages of small size,low cost and low-power consumption,but causes the disadvantage that the measurement system can only use fringes.Among these fringe-based structured light methods available,either the measurement accuracy is high but too many fringes cause slow measurement,or the measurement speed is fast but performance in robustness and measurement accuracy is bad.Difficulty of achieving satisfactory results in both measuring accuracy and speed is their main drawback.After an in-depth study on the above problem,we propose a reliable solution to significantly reducing fringe count at a minimal cost of accuracy,which greatly improves the measurement speed.The main contents are summarized as follows:1.Firstly,we build and calibrate a binocular three-dimensional measurement system composed of one MEMS chip and two infrared cameras in this project;besides,plenty of fringe images and ground truth data are obtained,and two error evaluation methods for disparity map and three-dimensional model are respectively given.2.Secondly,for the issue of fringe matching,a fringe structured light stereo matching algorithm based on ZNCC(Zero-mean Normalized Cross-Correlation)is proposed.After image preprocessing of face detection,noise reduction and epipolar correction,these fringe images are processed to obtain coarse matching results via the cross-correlation criterion ZNCC,then the consistency check and parabolic fitting are applied to refining the results.Experimental results demonstrate that this algorithm has high accuracy and can effectively solve the problem of fringe matching.3.Finally,for the issue of fringe generation,an optimal fringe structured light generation algorithm based on ZNCC is proposed.Here,the aforementioned matching algorithm is applied to the problem of fringe matching.Based on analysis of the relationship between fringes and matching errors,a theoretical optimization model of fringe generation in binocular vision is established,and then the generation formula of simulated fringe images and the solvable mathematical form of this model are derived after several assumptions are introduced.Generating train set and test set from previously obtained face data,stochastic gradient descent is performed on the optimization model for generating optimal fringes.Multiple simulations and actual experiments demonstrate that fewer optimal fringes can still achieve high measurement accuracy,and greatly improve measurement speed.