3-D Modeling from Video

"When humans move and observe, they combine information from the eyes and inner ears to form a mental world model using perception of linear and angular motion as well as visual images. We propose that recent advances in projective geometry, positionestimates from inertial and magnetic sensors, and rigid body kinematics can be fused to provide a powerful new way of effectively generating complete 3D models from video. Combining video data with inertial data provides a computer with the informationhumans use for non-contact observation of the world. Inertial data and camera kinematics can be used with the principles of projective geometry and optical flow to increase the number of image points that can be related across multiple views, increase theaccuracy with which the points can be related, identify views (video frames) that will efficiently produce a world model, allow software to combine world model fragments, and interpret and capture surfaces and regions as well as edges and vertices in acomplete world model. The bottleneck in the availability of ubiquitous 3D presentation and interaction materials is the acquisition of the 3D models from real world spaces and objects. If accurate, complete renditions of real spaces and objects can becreated with video sequences from one or more freely moving video cameras this bottleneck can be removed. Applications for these models can be found in the forest products industry for mapping timbe