Research On Mobile Manipulator Localization Based On SIFT Algorithm Combining Binocular And Monocular Visual

The research on mobile manipulator involves multisensor data fusion, motion control, path planning, localization, mapping and various other aspects. Among all the aspects, the localization method is the premise and basis for mobile manipulator to achieve other tasks. On one hand, vision sensor has a number of advantages over other sensors (sonar, laser, infrared, etc.) for it can provide large quantity of detection information, scan data rapidly, have a good robust characteristic; On the other hand, in a visual localization system, the binocular vision system will have an advantage over monocular vision system for its ability of gaining comprehensive information, high precise localization and stronger noise resistance. However, a binocular vision system has its limitation in respect of environmental view, which contributes to missing data in the environment regeneration of shaded and occluded areas. This paper focuses on mobile manipulator localization using a combination of monocular and binocular vision in unknown environments in view of the above situation. This main content can be outlined as follows:Firstly, it makes a generalization and summary to the current domestic and foreign research status and research method on mobile manipulator, and it points out several important and difficult problems while giving an introduction of relevant mobile manipulator localization technique. The research of the thesis is proposed to solve these problems, which has drawn its theoretic value and the practical meaning.Secondly, it introduces the hardware structure and control system of the mobile manipulator platform—MT-ARM, analyzes the coupling characteristic and the dynamical attributes of the mobile manipulator as well as establishes the kinematics model of differential drive that accepted a non-holonomic constraint in view of the difficulty of precise modeling for mobile manipulator which is a large complex nonlinear system with the consideration of the kinematical constraint of the mobile manipulator. It also describes the manipulation space of a four-DOF modular manipulator of MT-ARM, and then comes up with a kinematics model with joint angle variables by means of D-H (Denavit-Hartenberg) method. Test and simulation have been conducted to verify the correctness of the above method.Lastly, it conducts a research on measuring and localization with monocular and binocular vision combined basing on SIFT (Scale Invariant Feature Transform) algorithm, which targets on the problem that error may exist in shaded and occluded areas and inadequate luminance situations. In addition, with the algorithm of SIFT, features extraction and the stereo matching of the experimental environment, which is based on the combination of monocular and binocular vision is conducted. The feature map of the environment is established afterwards. On this basis, experiments of localization by means of binocular vision as well as the combination of monocular and binocular vision are conducted, and the errors are obtained by the comparison of the actual and given trajectories. The result manifests the dynamic performance and localization precision that the combination of monocular and binocular vision has offered.