A new method for passive location estimation from image sequence using adaptive extended Kalman filter

Using known camera motion to estimate the position of objects in a scene from an image sequence is an important problem in automatic aircraft navigation and robot vision. The most difficult issue associated with passive location estimation using sensor information is that the estimated results are very sensitive to the error of image co-ordinates and uncertainties of the sensor parameters. In order to enhance the robustness of position estimation, a novel position estimation algorithm, which we call the adaptive extended Kalman filter-based algorithm (AEKF), is developed. The conducted experiments demonstrate that position estimation result in the face of uncertainties of sensor parameters are much better than estimation based on the extended Kalman filter. The approach presented can be applied directly to the much broader problem of how to achieve good performance of the system with currently available limited-precision instruments