Development of a high-speed range sensor using structured light

Abstract

Range sensors based on structured-lighting techniques have been used successfully to obtain three-dimensional scene information. These systems make use of the principle of the optical triangulation to construct 2 1/2-D image. Finding the correspondences of scene features in the image is necessary to construct 3D information using the optical triangulation. Various methods are proposed to establish the correspondence. They required multiple images to solve the ambiguity of the correspondence problem. In dynamic environment, however, the scene changes during acquisition of the images required for the range measurement. To obtain measurement of a dynamic scene, a range sensor using only a few images or only one image is required. Near-degenerate setup of the camera and the projector is used to solve the correspondence problem with one grid pattern. The time cost to calculate triangulation is reduced by two lookup tables. The tables contain the information about the correspondence of the grid node and the epipolar line, the range measurement for each grid node and epipolar line pair, respectively. Due to the use of the tables, the whole measurement process is performed within 0.7s with Pentium133MHz. Most of the measurement time is spent on the image capturing and on the extraction of the stripe. Calibration of the projector is performed by space encoding. This method is easier and more accurate than the well-known two plane method which is used commonly in range sensors using stripe lines. The two plane method requires accurate movement of a calibration plane, but space encoding requires only the 3-D information of known points, i.e., the calibration marks on a calibration box. Our sensor can be used as a space encoding range finder, when only one projector is used. Thus, the comparison of the data accuracy can be performed very easily. Also, range measurements using grid projection and space encoding can be fused.