Research On Visual SLAM Technology Based On Multi-sensor Fusion

With the rapid development and breakthrough of science and technology in the field of automation,intelligent mobile and sensing robots are gradually active in the production and life of various industries.The autonomous positioning,navigation and sensing explo-ration of the surrounding environment of intelligent devices have become the hot spot and direction of mobile robots in the field of automation.However,when sudden disasters such as fires occur in indoor buildings without prior map information and GPS position-ing,mobile robots using traditional SLAM technology can move quickly and provide prior rescue information,which has problems such as easy tracking loss,large positioning error and time-consuming detection.These problems limit the robot to provide fast and accurate prior information for rescue work,which is not conducive to the rapid rescue work of the follow-up rescuers.Based on this,our paper will study the fast and accurate SLAM tech-nology and scheme of mobile and sensing robot in the scene of fire building.It is hoped to solve the difficulty of tracking loss and error increase under motion blur when using a single camera for visual SLAM,as well as the problem of large computation and long search time when using bag-of-word model to detect loop-closing.In order to overcome the visual failure of SLAM,reduce the error of SLAM and the computational amount and time of SLAM loop-closing detection,so as to provide rapid and accurate information support for the subsequent rescue work of fire personnel.In this paper,the SLAM technology and scheme required by robot in the fire scene are firstly discussed,and the SLAM technology and foundation required by robot in the disaster scene are pointed out.Put forward using visual SLAM,single binocular camera,inertial measurement unit and wi-fi sensor fusion scheme,solve fire scenario of mobile robot in the rapid movement and provide a priori information rescue the traditional SLAM technology tracking easy loss,the result error is big and time consuming problem and meet the target scenario robot in this paper,we study to better realize the SLAM function demand.Based on the above analysis of robot SLAM technology in fire scenarios,this paper designs a set of vision SLAM system scheme based on multi-sensor fusion to meet the functional support of mobile robots in sensing,positioning and mapping under the corre-sponding scene constraints.In this scheme,the shortcomings of feature extraction,pose estimation and loop detection in visual SLAM are analyzed and improved.In order to solve the problem that corner points in the front end of SLAM are connected and cannot fully reflect image information,a FAST corner algorithm based on adaptive grid and dy-namic threshold is proposed in this paper,which reduces the error of visual SLAM results by about 10% compared with ORB-SLAM2 scheme.In view of the difficulty of tracking loss and vision failure of SLAM schemes using a single camera under motion blur,a vi-sual inertia SLAM scheme based on factor graph fusion is designed in this paper,so that the trajectory tracking of SLAM is not easy to lose,and complete SLAM trajectory and results can still be obtained under fast motion.Aiming at the problems of large loop search space and long time of single detection caused by the use of bag-of-word model in loop-closing detection of SLAM,a loop-closing detection method based on Wi-Fi sensing is designed in this paper.The loop-closing candidate frame set selection and the evaluation of loop-closing detection are completed by using the Wi-Fi signal in the environment.The redundant computation in the loop back module of SLAM is reduced and the loop-closing detection is accomplished quickly.Finally,this paper verifies the vision SLAM scheme based on multi-sensor fusion,in-cluding ”Simulation Experiment of Open Data Set under Ubuntu16.04 System” and ”Mea-surement Experiment of Indoor Environment under Ubuntu16.04 System”,and verifies the SLAM scheme designed and implemented in this paper.The effectiveness and advan-tages of the design algorithm and scheme are verified.And the problems existing in the experimental results are discussed,the prospect of future SLAM research is proposed.