Research On SLAM Technology Of Indoor Mobile Robot Based On Vision-Inertial Fusion

With the continuous development of mobile robot technology,SLAM technology,one of its cores,has been continuously developed.The system dominated by laser sensors has been relatively improved,and the system dominated by vision sensors has been well promoted.With the continuous improvement of computer speed,the filtering solution method has been developed to the optimization solution method as the mainstream.At present,the visual SLAM system is not ideal for many problems in different scenes,such as the texturless environment,the violent shaking of the complex environment of the robot,and the closed elevator with movement.In this paper,both vision and inertial navigation sensor technologies are studied,with emphasis on the method of information fusion and optimization solution of visual inertial navigation,and a more robust and efficient visual inertial navigation SLAM system based on fusion of multi-sensor information is proposed.Experimental verification is also carried out.In this paper,the basic theoretical knowledge of mobile robot SLAM system is firstly studied,the essential problem of SLAM is described mathematically,and the motion representation of the robot in the coordinate system is described,which is mainly expressed by rotation and translation.Rotation involves rotation matrix,euler Angle,quaternion knowledge.The inertial navigation system and its attitude direction and velocity displacement are studied to understand its realization mode.Secondly,the existing pure visual SLAM system is studied,which involves camera model,camera calibration and other contents.The front end vision odometer uses the method of extracting the feature points of the ORB,carries out inter-frame matching,and solves the pose of the robot through PnP.Through the key frame data,the back end uses the local BA method to optimize,and adjusts the local pose and map points.At the same time,the loop detection based on the word bag model is added to realize the global optimization.Then,aiming at the shortcomings of the previous pure visual SLAM system,a visual inertial information fusion SLAM system was proposed.The whole process is as follows:for the information of different frequencies(such as vision 30FPS and inertial navigation 100HZ),the IMU information is pre-integrated,and then solved with the visual information through tight coupling;The back end realizes local optimization through the initial pose obtained by the front end tight coupling and the existing local map points.The global optimum is realized by eliminating the accumulative error by loop detection.Finally,the experiment was carried out on the data set and the actual hardware and software platform,and compared with other SLAM systems.The experimental results show that the SLAM system with visual inertial navigation in this paper has good robustness and efficiency on the indoor mobile robot platform.