A Novel Approach Of An Fpga Design To Improve Monocular Slam Feature State Covariance Matrix Computation

Monocular SLAM is a study which concentrates on deriving the position and motion estimation information from tracked features using a single camera. In this paper, a novel approach to improve the computation speed of a Monocular SLAM is proposed. The research concentrates on the feature initialization process which takes place before the standard Extended Kalman Filter (EKF). In order to find the most time consuming process at the initialization stage, a software profiling tool is used. From the result, the section of a program which demands high processing computation is identified. Following that, a specialized design is proposed to improve the computation speed. An FPGA approach is chosen with the intention to offload software processing to a dedicated hardware for overall performance acceleration. In order to accomplish this goal, the section demanding high processing computation is carefully studied. From the studies, it is found that the original approach can be improved by reducing the multiplication process and incorporating parallel processing capability of an FPGA. At the end of the paper, the comparison results of the software and hardware processing are presented.