Visual SLAM using only images as external information estimates the robot position while building the environment map. SLAM is a basic prerequisite for autonomous robots. Now it has been solved by using a laser or sonar sensor to build 2D map in a small dynamic environment. However, in a dynamic, wide range and complex environment there are still problems to be solved, and the use of vision as the basic external sensor is a new area of research. The use of computer vision techniques in visual SLAM, such as feature detection, characterization, feature matching, image recognition and recovery, has still much room for improvement. The paper offers a brief overview on visual SLAM about the latest and easy to understand technologies in the field. Multi-robot systems have many advantages over a single robot, which can improve the precision of SLAM system, and better adapt to the dynamic and complex environment. This paper expounds the methods of multi-robot SLAM, with emphasis on the map fusion methods.

multi-robots  /  SLAM  /  computer vision  /  map merging