Enhanced Simultaneous Localization and Mapping Keypoint Detection through Deep Learning Methods

This research focuses on a novel approach to Simultaneous Localization and Mapping (SLAM) for autonomous mobile robots. SLAM enables the robot to move independently in a geofenced environment by building a dynamic map of its surroundings. The current methods of SLAM are computationally expensive due to the density of keypoints needed to localize the robot correctly, which limits the robot’s autonomy. Keypoints are also inefficiently recalculated on a per-frame basis, with the number and location of keypoints not necessarily assigned the same way as the previous frame. This research aims to create an enhanced keypoint detection and processing system using a deep learning model. The focus is on limiting the number of keypoints, typically in the thousands for textured surfaces, to a few hundred. By using semantic segmentation, the keypoints bind effectively to the object perimeter, giving the robot the ability to perform a broader range of tasks in various environments.