Improved Visual SLAM Using Semantic Segmentation and Layout Estimation

The technological advances in computational systems have enabled very complex computer vision and machine learning approaches to perform efficiently accurately. These new can be considered a set of tools reshape the visual SLAM solutions. We present an investigation latest neuroscientific research that explains how human brain accurately navigate map unknown environments. accuracy suggests navigation is not affected by traditional odometry drifts resulting from tracking features. It utilises geometrical structures surrounding objects within navigated space. identified space shapes anchor estimated representation mitigate overall drift. Inspired brain’s techniques, this paper presents our efforts incorporate two techniques into VSLAM solution: semantic segmentation layout estimation imitate abilities proposed system benefits relations between corner points cuboid environments improve trajectory estimation. Moreover, implemented solution semantically groups then tracks each group independently limit yielded higher immunity large pure rotations.