Visual SLAM for Mobile Robot in a dynamic environment

-> Get full project report Here <-

Robot localization is an integral part in mobile robotics. It is the base for path planning and navigation tasks for robots and for AR/VR applications as well. SLAM has been a well known method for mapping the unknown environment and localizing yourself in the map. Visual SLAM, the category of SLAM, makes the use of visual sensors such as cameras to perform SLAM. Such visual sensors are available at cheap cost nowadays and hence it is one of the most researched and popular topics in mobile robotics.

This project uses a camera as its only sensor to build a 3d map of the entire room and localize itself in the built map. The map can then be used for navigation purposes within the mapped environment. Problems such as dynamically changing environment, varying lighting conditions, lack of textured environment are the hindrances for visual SLAM. Some of these problems have been well tackled in this project. Dynamic objects in the environment have been masked to minimize its effect. Light invariant feature extraction has been used to tackle variations in lightning conditions.

The Robot Operating System (ROS) has been used to communicate between various parallel processes and between robot and PC. Most of the high computational tasks have been done on the PC. The robot then responds on the basis of commands given by processes which are running on PC.