Research On Simultaneous Localization And Mapping Of Duct-cleaning Robot

With the developing of technology and society, central air-conditioning has been widely applied in the public environment we live,we work and others. As the central air-conditioning runs all year round, variety of harmful substances deposited in the pipeline, which flows into rooms together with the airflow, seriously influenced the indoor air quality. But the central air-conditioning pipelines crisscross complexly, artificial cleaning is not only time-consuming, low efficiency, but also unsafe to human nature. So the research and development of air-duct cleaning robot to accomplish the task of cleaning has important research values and urgent practical significance. However, in order to realize the self cleaning of the air-duct robot in the central air conditioning system, the simultaneous localization and map building (SLAM) of the robot is a key technology that must be solved. In this paper, a method of simultaneous localization and map building for a robot with six dimensional inertial measurement units and a binocular stereo vision sensor is proposed. The main work in this paper can be summarized as follows:(a) The working environment and working principle of the air-duct robot are introduced, and the overall structure and the function of the module are analyzed. In this paper, the perspective projection model of camera and the fundamental principle of binocular stereo vision are introduced, and basic system structure of air-duct robot binocular stereo vision is proposed.(b) In this paper, the algorithms of SIFT, SURF, ORB feature extraction are introduced, and their relative performance is compared with experiments. The ORB is superior to the SIFT and SURF, since matching speed is improved more than an order of magnitude under the similar matching accuracy.The matching strategy of ORB algorithm is optimized, and the matching speed is improved. In order to eliminate the false matching, The matching results are optimized by PROSAC algorithm,and the matching accuracy is improved.(c) The classical FastSLAM algorithm were briefly described, and some problems of its solution is proposed. Firstly, the transformed cubature kalman filter is used to calculated the SLAM posterior probability density in order to improve SLAM accuracy and reduce linearization error; Instead of covariance matrixes, the covariance square root factors is propagated in SLAM process to avoid the time expensive Cholesky decompositions and improve computation efficiency. Secondly, Focusing on particle impoverishment of the traditional FastSLAM, a improved Cuckoo Search (ICS) is used to optimize the proposal distribution. ICS causes the particle set to approach to high probability region of the posterior and makes them distributed closely to the true pose. Finally,under the simulation and experimental data sets Sydney feasible, the algorithm is applied to the air-duct robot binocular vision system, and experiment were carried out with good result.