Hybrid Feature Based SLAM Prototype

TL;DR

This paper presents a Visual SLAM-based navigation system for robots that uses stereo vision and Fast SLAM techniques to accurately recognize features and build 3D maps in indoor environments, tested with prerecorded video.

Contribution

It develops a Visual SLAM framework combining Fast SLAM with stereo vision and feature recognition, implemented in MATLAB for indoor robot navigation.

Findings

Accurate feature recognition and data association in stereo images.

Successful 3D mapping and localization in indoor environments.

Effective implementation of Fast SLAM with stereo vision in MATLAB.

Abstract

The development of data innovation as of late and the expanded limit, has permitted the acquaintance of artificial vision connected with SLAM, offering ascend to what is known as Visual SLAM. The objective of this paper is to build up a route framework dependent on Visual SLAM to get a robot to a fundamental and new condition, have the capacity to set and make a three-dimensional guide thereof, utilizing just as sources of info recording your way with a stereo vision camera. The consequence of this analysis is that the framework Visual SLAM together with the combination of Fast SLAM (combination of kalman with particulate filter and SIFT) perceive and recognize characteristic points in images so adequately exact and unambiguous. This framework uses MATLAB, since its adaptability and comfort for performing a wide range of tests. The program has been tested by inserting a prerecorded video input with a camera stereo in which a course is done by an office environment. The algorithm initially locates points of interest in a stereo frame captured by the camera. These will be located in 3D and they associate an identification descriptor. In the next frame, the camera likewise identified points of interest and it will be compared which of them have been previously detected by comparing their descriptors. This process is known as "data association" and its successful completion is fundamental to the SLAM algorithm. The position data of the robot and points interest stored in data structures known as "particles" that evolve independently. Its management is very important for the proper functioning of the algorithm Fast SLAM. The results are found to be satisfactory.