Ill-posedness analysis and robust estimation for multi-sensor spatial registration

In multi-sensor spatial registration process, the ill-posedness of the information matrix is the main factor which affects the quality of the registration result. This paper focuses on this problem and makes some progress. For simplicity and without loss of generality, we firstly revisit the least square registration algorithm and analyze the ill-posedness of the information matrix and its reasons. To solve the ill-posed problem, a novel matrix singular value modification algorithm, i.e., the Combination Singular Value Decomposition (CSVD) algorithm is proposed by using the trapezoidal distribution characteristic of singular values of Fisher information matrix. The main contribution of the proposed CSVD algorithm is that it can fully preserve the determinate part of information matrix, while effectively suppressing the uncertain part. Finally, simulation results of two different scenarios show that, the proposed CSVD method can significantly improve the stability and accuracy of the sensor spatial bias estimation.