Estimation of 3D object points from omni-directional images acquired by a rotating line camera

As various service models integrating 3D GIS with augmented reality are being developed, the demand for indoor spatial data has been increasing rapidly. These data can be effectively derived from indoor omni-directional images, which can be acquired by a stereo rotating line camera. In this study, we thus develop a method to estimate 3D object points from multiple overlapped omni-directional images acquired by the rotating line camera to construct indoor 3D models. Based on the geometry of the rotating line camera, we derive modified collinear equations relating an image point in an omni-directional image with its corresponding object point. Based on these new collinear equations, we estimate the 3D coordinates of object points appearing in at least two images. In the experiment, we acquired two omni-directional images by the rotating line cameras with different rotation radius and applied the proposed method to these data. The result shows that the 3D object points can be obtained within ±15 cm accuracy. In future, we plan to develop the bundle adjustment for a rotating line camera and estimate the exterior orientation parameters of each omni-directional image as well as the 3D coordinates of object points. The results will then be used further for indoor 3D modeling.