Dual Affine Transform Method and Extended Kalman Filter for Target Extraction and 3-D Position Location from Video Images

We propose to develop a dual affine transform method and an extended Kalman filter for extracting target from video images, and locating its three dimensional (3-D) position in GPS coordinates. Due to the changing perspective, the appearances of the selected target will be different. Rapid motion/geometric compensation is essential to target extraction. Also, in order to estimate camera motion, it is necessary to obtain accurate two dimensional geometric/kinetic measurement for the background in images. Dual affine transform method offers an excellent candidate for the sophisticate motion compensation and image 2-D geometric measurements. Using 3-D rigid motion and perspective projection models, and with the camera positions established by GPS and inertial navigator, extended Kalman filter scheme can be employed to estimate target 3-D position from the 2-D geometric measurements. Compared with constrained inverse problem formulation currently being used in computer vision, the extended Kalman filter has the advantage of incrementally improving the accuracyof 3-D position by exploiting high redundancy of 2-D measurements in the sequence of video images. Phase I work will develop robust algorithms to meet navy's requirement of accuracy under operation constraints and show the feasibility of our approach to video image target extraction and location by demonstrating its performance.