Object Extraction Method for Mobile Robots using Fast Growing Neural Gas

In this work, we present a fast object extraction method that can be applied to embedded computers in mobile robots. Currently, perceiving the real-world in 3D space is becoming increasingly important, and robotics is no exception. Many types of robots have utilized 3D point cloud data as part of their mission, such as rescue robots, humanoid robots, explorer robots, etc. All of them make use of 3D sensor data as the input. However, a problem occurs when large amounts of data are generated, especially for embedded computers that are usually installed in mobile robots. Therefore, we propose an object extraction method for 3D point cloud data that is lightweight and can be applied to the embedded computers. It is developed based on a simplification of fast growing neural gas algorithm, called as simplified FastGNG. As opposed to the standard GNG algorithm, FastGNG uses a multi-scale batch learning strategy to enhance the learning speed. Thus, in this work, we applied the algorithm to extract information of the point could data as a way to compress the large amounts of data. The experimental results have shown that the proposed method is able to reach 36.2 fps which is real-time capable, and hence suitable for embedded computers.