ORB-SLAM With Near-Infrared Images and Optical Flow Data

Antonio Buemi, Arcangelo Bruna, Sylvain Petinot, Nicolas Roux; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, 2021, pp. 1799-1804

The algorithms designed to solve the Simultaneous Localization And Mapping (SLAM) problem have to be often executed on embedded platforms in order to become part of complex robotics systems. Despite the continuous growth of their computational capabilities, the embedded devices still have considerable limitations, especially in terms of memory. This paper presents a modified version of the well known ORB-SLAM algorithm which improves its performance thanks to the use of Hardware-generated Optical Flow (HW-OF). The ORB-SLAM has been modified in order to run into the Stereo-cam embedded system by STMicroelectronics. The Stereo-cam includes the VD56G3 sensor, able to provide Near Infrared (NIR) images and OF data computed by a hardware accelerator. The experiments showed an improvement of the ORB-SLAM performances in terms of memory consumption and frame rate.