Surface Structure Light Subpixel Cloud Point Extraction And The Optimization Method Of Two Step Calibration Study

Structured light technology is an important research field of computer vision.Based on binocular vision, the structure light technology utilize a projector that canproject the mode with coding information to target scene instead of one camera inbinocular stereovision. Through the decoding and analysis to the images that capturedby the second camera, we get the depth information, so as to realize the reconstructionof the3d scene. The uniqueness of model coding can effectively solve the problemthat the binocular vision technology hard to match the pixels and has the advantagesof simple, efficient, etc. So the computer vision based on the structured light hasbecome the investigative hotspot for researcher and has been applied widely in manyindustrial and agricultural fields such as industrial automation, object recognition,robot navigation and object tracking.One of the key of the structured light detection technology is calibration. In orderto get a high precision detection, we should ensure that the structure light cloud pointsbe extracted with high precision and system calibration is high precision. Projectgroup proposed a structured light three-dimension detecting approach based onM-arrays, which aimed at dynamic three-dimension scene reconstruction using binarycoded light pattern. In view of the deficiency of the early work group, this paperimprove and optimize the two steps calibration method through researching the keytechnologies of the sub-pixel extracting of cloud points and the calibration methodwith the help of computer. The main research contents are as follows:（1）The study on sub-pixel extracting of cloud point. First, three kinds ofsub-pixel location are analysed and researched, and the advantages and disadvantagesof this three methods are summarized. Then the basic flow of extracting the cloudpoint sub-pixel location are established using threshold segmentation, connecteddomain extraction, the curve fitting, and the angular point extracting method. Andthen the extracting steps of the article light centerline are introduced. Aim at solvingthe problem that the segmentation using only one threshold can’t solve theinterference of the light to the segmentation, the block threshold segmentation method has put forward. This method once can complete image threshold segmentation withgood segmentation effect. This paper also put forward a method that can automaticjudgment gray mutations position to select fitting sampling points with better curvefitting effect, and more accurate position of the center point of the light line. Throughthe fix to the individual center points data of light line, the noise interference has beenruled out. The experimental results show that the algorithm of this system promotedthe accuracy from pixel level to sub-pixel level, and have higher extraction accuracy.（2）The study on system calibration. The camera calibration foundationknowledge and the traditional calibration method have been studied. In view of theproblem that using many times artificial measure to match the position relationsbetween the experiment platform and calibration surface will cause larger error in fasttwo-steps calibration method in our subject group’s early work, a new method hasbeen put forward that adjust the position relationship between the experimentplatform and calibration surface in the help of computer. This method gives thecapability to the computer that can automatic judge the position relationship betweenplatform and calibration surface and show out the data in the calibration interface. Weadjust the platform according to the information until adjust successfully. Theexperimental results show that this method provides higher calibration precision thanthe method in early work. Experiment results show that the relative error of thesystem calibration is about one sixth of the error of the early work, calibrationprecision is distinctly improved.Key technologies of the computer vision system such as the sub-pixel localizationof the image points, system calibration, the calculation of three-dimension coordinateshave been studied. At the same time, the software has been developed. The results ofexperiment show that the methods used in this paper are correct and effective. Thiswork provides a foundation for developing and constructing a while computer visionsystem.