An Improved Pose Estimation Algorithm For Real-time Vision Applications

Abstract

An accurate and efficient pose estimation algorithm is presented to estimate the location of a three-dimensional object from the relating two-dimensional images. The iterative pose estimation algorithm, called OI (orthogonal iteration) algorithm, is fast and globally convergent. But it seems that its simulation results have a large translation error in relatively close range, which was explained as the influences of biasing effects introduced by the projection matrix; however, we find that it was caused by the error of rotational matrix. Based on improving the accuracy of rotational matrix, the OI algorithm was refined. Experiments with both synthetic data and real images show that the results using the improved method are more accurate than that of the OI algorithm