Application of Robust Regression for Exterior Orientation of Video Images

Nowadays, the 3-D measurement system using the video image sequences has been used in many fields, for example the vision sensor for machine control or the measurement technique of city area for GIS or landscape simulation. However, the robust estimation of the exterior orientation parameters for each video frames is still important issues. In recent years, the authors have been concentrating for development of 3D measurement system using video image sequences with robust tracking and robust exterior orientation. However, some problems should be resolved for this systems goal, such as limitation for camera movement, robustness of bundle adjustment. Therefore, the authors are developing the new algorithm of robust exterior orientation for video image sequences using robust regression procedure. In this paper, authors describe the effectiveness of robust exterior orientation method for video image sequences which developed by authors. Moreover, application for 3D measurement of this algorithm is also described. 1. INTORDUCTION In order to perform the 3D measurement using video image sequences from the platform moving freely, the technique of exterior orientation for each video frame is still important issue. In the field of general photogrammetry, the exterior orientation is performed using many control points that have accurate 3D coordinate. In addition, the position and rotation sensor such as GPS or gyro sensors are used in the real-time application field of 3D measurement from video image sequences recently. However, the construction of accurate control point generally needs a lot of time and labor. In the case of GPS, the accuracy of positioning is depending on GPS satellite position and condition of the environment, and the positioning at inside of structure is particularly impossible. Similarly, the high accurate gyro sensor’s cost is still expensive. In addition, high accurate synchronization in time axis between these sensors and video camera is necessary, and it becomes the rise in size or cost of system. From these backgrounds, automatic estimation technique of the exterior orientation parameters from video image sequences without using the above-mentioned equipments has become important issue in late years, and many research groups had shown the studies and applications [1, 2, . On the other hand, the authors have developed PC-based 3D Image Measuring Station called PI-3000, and many applications of 3D measurement using consumer digital still cameras have been achieved previously. This system can estimate the interior orientation parameters of consumer digital still camera accurately. Furthermore, this system can perform bundle adjustment with many images simultaneously. Therefore, exterior orientation parameter of camera and 3D coordinate of object can be obtained in high accuracy [4, . From circumstances mentioned above, the authors have been concentrating for development of the 3D measurement system using consumer video camera with robust tracking and robust exterior orientation method. In this method, in order to perform the automatic rejection for error correspondences of natural feature points in each video frames, tracking process and relative orientation process in exterior orientation procedure performed robust regression based on the LMedS (Least Median of Square) method. However, in this exterior orientation procedure, there is a limitation for movement of a camera because of the limit of relative orientation procedure. In addition, the robustness of relative orientation is insufficient when there is the big change of the scene such as the structure corner . Therefore, in order to resolve these problems, the authors are developing the new algorithm of robust exterior orientation for video image sequences. In this paper, authors describe the effectiveness of algorithms of robust exterior orientation for video image sequences, and application for volume measurement from video image sequences is also described. 2. EXTERIOR ORIENTATION PROCEDURE FROM VIDEO IMGAE SEAQUENCES 2.1 Main Flow of Exterior Orientation Procedure The exterior orientation procedure in this system consists of the tracking process and bundle adjustment process. The tracking process performs the extraction of natural feature points from video image sequences and tracking the natural feature points. As the result of tracking process, corresponding points in each video frames are obtained. Finally, bundle adjustment for all