Research On Mobile Robot SLAM Algorithm Based On Vision

The simultaneous localization and mapping is one of the most important techniques for making a robot truly autonomous in localization and navigation.After a long period of development,the SLAM technology has formed in two kinds of methods based on filter and nonlinear optimization.In recent years,with the application of consumer cameras and the development of computer image processing technology,the SLAM using vision methods has drawn more and more attention,and has gradually become an important branch in the research field.Additional,visual SLAM has a great potential to be implemented into the mobile robot field.Considering the bright future of SLAM,the research which begins at the filter based method puts focuses on the vision based SLAM technology.The SLAM algorithm based on the traditional EKF method has some shortcomings in terms of the positioning accuracy and robustness due to the linearization error and the error accumulation of process noise.To improve this problem,adaptive fading factor has been adapted into the system,resulting in a new SLAM method.Moreover,the simulation experiment and the test with the open source data have been carried out,and the results show that the algorithm has been improved in aspects of positioning accuracy and robustness.On this basis,the mono vision SLAM has been studied.Analysis of the extraction and matching method of feature points which is adapted in the 1-point-RANSAC vision SLAM has been carried out.Meanwhile,the inverse depth parameterization has been introduced to initialize the new map points.Besides,the adaptive fading EKF method has also been applied into the mono vision SLAM,with experiments carried out at indoor environment.To cope with the problem that mono camera can't get the environment depth information directly,which would cause scale drift,stereo camera has been implemented to fulfill the requirement of SLAM.Therefore,the principle of binocular parallax distance measurement has been analyzed.At the same time,the image features extraction methods have been compared.Additional,the stereo vision SLAM system has been studied completely.Meanwhile,the algorithm has been tested with the challenging open source data set,achieving an excellent result in localization accuracy and system robustness.