Autonomous Topographic Mapping of Unknown Environments by Dynamic Visual Data

Abstract

The simultaneous localization and mapping (SLAM) process is what makes it possible for autonomous vehicles to navigate in unknown environments. Early SLAM algorithms used and relied on ranging sensors only. In recent years, there has been an increased interest in vision-based SLAM (V-SLAM) due to the low-cost nature of digital cameras available in the market compared to ranging sensors. V-SLAM uses successive camera frames to either track features in individual frames and triangulates the position to construct a 3-D map or determine the vehicle speed by measuring the rate of change of these features relative to a known reference. This paper proposes an effective real-time method of creating a topological 3D map of the environment from a stereo vision system by using an improved stereo correspondence algorithm that minimizes errors caused by illumination and texture variation in the disparity map generation. The Cartesian world coordinates corresponding to each pixel are computed from the disparity map generated by triangulating the depth of the pixels in the reference perspective projection image to create a 3-D map of the scene as a point cloud plot. Analysis of the resulting point cloud plot indicates that the coordinates of each pixel provide the 3-D information about the scene representing a working topological map that can be used to detect the obstacles in close vicinity to the robot.