Extraction Of Point Cloud Normal Based On The Line Structured Light Image In Natural Environment

Line structured light measurement has been widely applied in3D measurement anddetection because of its many excellent properties, such as low cost, small volume,lightweight property, ease and flexibility. In the process of line structured light measurement,when the surface curvature of the object changes dramatically, the normal of the originalcenter must be calculated firstly, and then the light strip center can be obtained accuratelyfrom the direction of the normal. Therefore, studying the effective and accurate methods toextract the point cloud normal has become a priority and hotspot issue in the research of linestructured light measurement. In this paper, we study and realize a high-precision andhigh-speed point cloud normal extraction method. Our main research content and conclusionare as follows:(1) We select the wavelength of laser and solve the precise distinction problem betweenlaser and natural light in CCD/CMOS images. Firstly we calculate the density of solar energyby Planck formula,and consider the factor of atmospheric absorption.Then we can obtain thesolar spectrum curve. Next we choose the appropriate laser wavelength through thecombination of the solar spectrum curve and laser safety standards. At last, we get theresponse coefficient by multiplying the values of corresponding wavelength in solar spectrumcurve and sensitive curve. We also design the laser scanning device which can be used innatural environment.(2) We extract the initial value of the light stripe center. In actual measurement, the linestructured light image is always affected by the random and non-random noise, in order toavoid these problems, some preprocessing operations including gray processing, imagefiltering and threshold segmentation must be conducted to the light stripe images. Thenconsidering the current status of light strip center extraction algorithms, initial centers areextracted by Hildith refinement algorithm and barycenter method. Finally the extractionresults of the two methods are given and compared. The results show that the Hildithrefinement method will lose some points, so we select the barycenter method to acquire the point cloud.(3) We propose a circumcircle method to extract the point cloud normal, and it cansatisfy the requirement of both accuracy and efficiency. After obtaining the initial light stripcenters, we need to judge the collinearity of every three-adjacent point firstly, and if it doesn’tsatisfy the collinearity property, the center coordinates of its circumcircle are calculated. Theinitial normal of every point is the connection line with its center coordinates of itscircumcircle. At last, in order to reduce the concussion, we average every three-adjacentnormal.Experiments of our method on three different light stripes show that the efficiency isincreased by3times when compared with Hessian matrix method, and also doubles that ofSobel gradient method. In addition, the average error is less than0.01mm. So thecircumcircle extraction method can satisfy the requirements of line structured light3Dmeasurement.