| The autonomous navigation of mobile robots is one of the key technologies in the field of robots,and navigation can generally be decomposed into two modules: positioning and path planning.The main research direction of this dissertation is based on the information fusion autonomous positioning technology of multi-sensors such as vision,inertia,odometry,and so on.The application background is mainly for indoor and other non-satellite signal sites.The inertial system has become the core positioning equipment of the mobile robot with its characteristics of independence,stability and all-weather service.However,due to its inherent flaws in positioning error accumulating over time,it's not satisfied the needs of long-haul navigation.With the development of computer and other information technologies,visual SLAM has become a research hotspot in the field of positioning and navigation.There is a clear complementarity between vision and inertia.Their combination can effectively improve the accuracy and stability of navigation systems.This thesis builds a multi-sensor combined positioning framework based on ROS system,focusing on the improvement of visual SLAM,visual inertial odometer combination model and so on.In this paper,the algorithm is transplanted into the embedded platform,besides simulation experiments and prototype experiments are used to verify the effectiveness of the algorithm.The specific research work and achievements of this paper are as follows:Firstly,the basic principle of strapdown inertial navigation is studied.The quaternion is used to update the attitude matrix,and the dead reckoning method is used to update the speed and position.For the error of inertial navigation,the inertial odometer model is established in this paper to correct the INS information in one step,and the correctness of the model is verified by simulation.Secondly,the basic theory of visual SLAM in rigid body motion estimation in threedimensional space is studied.For the problem that the visual feature point matching error rate is high,this paper proposes a clustering sampling consensus algorithm based on image priori for improvement.For the matching error in loop detection,this paper proposes a multi-scale visual pouch method based on key frames,which has a strict detection of the loopback accuracy.Through simulation experiments of different data sets,the effectiveness of the improved visual SLAM method is verified.Thirdly,for the case that the RGB-D camera has unknown depth information of some pixels in the actual situation,the ICP and PnP algorithms are mixed and the re-projection error function is re-derived.Then the Lie algebra is used to solve the perturbation term of the error and the error function is optimized.Fourth,a combined positioning framework for vision and inertial odometers is established,in which the self-calibration error state quantities including installation errors and coordinate system drift were added.In addition,according to the background of actual navigation,a method of multi-sensor fusion data synchronization is proposed,and fault diagnosis of visual pose estimation is made.Finally,through the data set simulation experiment,the superiority of the combined model relative to pure visual positioning is verified.Fifth,based on the ROS system,the software algorithm proposed in the paper is transplanted.On the hardware platform,the RGB-D camera driver is transplanted.Based on the hardware and software platform,the accuracy of the visual SLAM algorithm and multisensor information fusion location algorithm in the paper is verified by prototype experiments,and the data transmission between the positioning system and the route planning system was completed,which provided a basis for autonomous navigation. |
PDF Full Text Request