Reducing multiplication operation and independent processing for monocular simultaneous localization and mapping (SLAM) feature state covariance matrix computation

Abstract

Monocular simultaneous localization and mapping (SLAM) research is a study which concentrates on how to derive position and motion estimates information from tracked features using a single camera. Before the features can be processed by standard extended Kalman filter (EKF), they have to be initialized. In the initialization process, the state covariance matrix calculation is found to be the most time consuming process. This is proven by software profiling method which is used to identify which section of program demand high processing computation. The execution time is further increased when the number of features is increased. This is due to the fact that the matrix multiplication involved in obtaining the state covariance becomes larger when more features are added. In this paper, the author proposed a new method to reduce the computation time by altering the state covariance matrix formula by reducing the multiplication operation involved. The proposed method also manipulates the conventional approach to produce multiplication process which is independent. The independency will enable future researcher to consider parallel design which would further accelerate the execution time. Key words: Simultaneous localization and mapping (SLAM), parallel design, matrix multiplication, landmark initialization, inverse depth parameterization