Motion and Structure using Multi Estimation Technique

摘要

A novel approach for Structure from Motion (SFM) problem has been suggested. A recursive framework has been used to find error in measurement of motion and structure parameters. Using feature correspondence it has been suggested that instead of using a single Extended Kalman Filter (EKF) to estimate the state vector an Interactive Multiple Model (IMM) Filter can be used which divides the state vector and estimates it in two steps. A set of three EKFs have been used to track the motion of an object and another set of N EKFs have been used to find structure parameters. The whole arrangement of two stage filters works in an interleaved manner. The first stage estimates motion parameters while second stage estimates structure. The proposed scheme requires 3 + N small EKFs in contrast to N x N used by other researchers. This reduces time complexity from quadratic to linear compared to approaches which use a single full covariance EKF. The technique results in lower error than existing algorithms.