Study on the strategy of 3D images based reconstruction using a camera attached to a robot arm

The metric distance error between the calculated point and the real point in the 3D measurement coordinate system is due to optimization methods carried out in 2D image plane. Euclidian image based 3D reconstruction is carried out in three major steps, geometric primitive extraction, correspondence, and triangulation. Extraction and triangulation are purely geometric tasks but the correspondence step is a challenge in precision. In this paper we study 3D object reconstruction based on a set of 2D images. We used a robot arm to accurately move the camera and we study the influence of the motion. It has been demonstrated that the impact of a correspondence error on the reconstruction accuracy of a 3D images-based reconstructed point may vary depending on the image capture strategy. Two modes of motion have been adopted. The first is an axial mode making the camera close or far from the object acting like a zoom. The second mode is lateral where the camera moves parallel to the object. It has been shown that axial mode leads to greater errors compared to the lateral mode. The impact of the geometric parameters of photographing on images based 3D reconstruction error is also discussed.