Bionic Visual-based Data Conversion for SLAM

Abstract

Simultaneous localization and mapping (SLAM) is the key function for most mobile robots to achieve autonomous navigation. The traditional visual SLAM uses the camera to acquire data and constructs a sparse or dense 3D map, which is convenient for robot localization but difficult for obstacle avoidance and autonomous navigation. Thus, we propose an innovative data conversion algorithm based on bionic visual characteristics which can construct a two-dimensional accurate map for indoor navigation in this paper. The algorithm has two main parallel threads: Ground Detection and Data Conversion. The ground detection thread detects the ground in real time, and gets the transformation matrix from the camera to the ground based on the geometrical invariability. The data conversion thread first filters the depth data, and then proposes a variable-resolution model based on human visual characteristics, which can keep the conversion time consumption at a low level without affecting the accuracy. Each group of experiments shows that the data converted by our algorithm have high-precision, and can be used to construct the map for navigation accurately.