Research On Simultaneous Localization And Mapping Based On Stereo Vision

The technology on robotics is developing rapidly. As one kind of the basic technology,Simultaneous Localization and Mapping are always a popular research point amongrobotics researchers. The development of sensor industries has significant influence onthe development of robotics for the sensors carried by robots decide how the robotperceive and describe the outside world. As Microsoft releases a new kind of depthcamera, Kinect, in a very low price, robot can get the depth information of theenvironment detected at a very low cost which bring a new round of upsurge of roboticsresearch. In this paper, we review the development of the technology on roboticssimultaneous localization and mapping, analyze Kinect camera and compare differentalgorithms on solving SLAM problems. Then we propose a new way to create a virtuallaser scanner based on RGBD data from Kinect and prove it is better than the old ones.We also build a simulation environment using Robot Operating System and Gazebowhich we use to verify the algorithms we propose. The simulation experiment shows thatthe virtual laser scanner can works more like a real one. In the second half of this paper,we analyze some problems exists in the Particle Filter algorithm used to solve the SLAMproblems. We analyze the algorithm step by step and find why the time complexity of itis so high. And we finally propose two ways to reduce the complexity of PF algorithm:one is to use less particles which will reduce the complexity of each weighted resamplingstep, and another is to reduce the time of weighted resampling step which will reduce thecomplexity of the PF algorithm as a whole. At last, we show experiments we do to provethe improvement is feasible and efficient.