The fastest visual ego-motion algorithm in the west

Ego-motion consists in estimating a camera motion relative to a rigid scene. It is a fundamental task for auto-guided vehicles or self-location of mobile robots. Unfortunately, recent works have demonstrated that computational complexity of ego-motion algorithms is a challenge for embedded systems because there are iterative operations that limit the processing speed. In this work, we propose a hardware architecture for the ego-motion challenge that consists of using look-up tables and a new feature matching algorithm. Unlike previous work, camera motion is estimated with no iterative loop and no geometrical constraints. As a result, the algorithm complexity is lower than previous formulations and experimental results demonstrated that a small size GPU-based implementation is feasible and suitable for embedded applications. Experimental results demonstrated that it is possible to reach higher accuracy (95.07% level of confidence) compared with previous monocular visual odometry algorithms such as CNN and depth learning-based approaches with a processing speed up to 17 times faster than previous works. (C) 2019 Elsevier B.V. All rights reserved.