Map Construction And Robust Localization Method Based On Multi-sensor Fusion

In recent decades,with the continuous development of autonomous mobile robot technology,Simultaneous Localization and Mapping(SLAM)technology has been widely concerned.At present,according to the types of sensors used,slam technology can be divided into two directions: vision and laser.Among them,laser slam has been applied in practice due to its high stability and accuracy,but there are still some problems,such as high re positioning time consumption,low positioning and mapping accuracy in dynamic environment.Aiming at the problems of high failure rate of loop detection,low positioning accuracy and slow relocation speed in indoor autonomous mobile robot mapping,this paper proposes a map construction and robust positioning method based on multi-sensor fusion.The main research contents are as follows:Aiming at the problem of low accuracy of wheel speed odometer motion estimation based on single sensor,an extended Kalman filter algorithm(EKF algorithm)is proposed,which combines the wheel speed odometer based on encoder(hereinafter referred to as wheel speed odometer)and inertial measurement unit(IMU)motion estimation method,that is,the extended Kalman filter first uses the wheel speed odometer measurement data to update the predicted value to the posterior value,and then uses the IMU measurement data to update the posterior value to the final posterior value.The experimental results show that the motion estimation method based on EKF algorithm fusing wheel speedometer and IMU has higher accuracy than that based on wheel speedometer.Aiming at the problem of high failure rate of loop detection caused by environmental interference,a method of loop detection based on visual assistance is proposed.Firstly,the current frame is matched with all the previous frames to narrow the detection range of loop,and then the loop detection method based on branch and bound and correlative scan matching is used to traverse the sub images of similar frames to get the loop information.Experiments show that the feature matching module based on visual bag of words model can improve the robustness of loop detection.In order to solve the problem of high time-consuming relocation due to the lack of prior pose information,a relocation method based on laser feature matching is proposed.Firstly,the laser feature points of all the past laser frames are extracted and clustered by two detectors to establish the laser bag library,and then the similarity of the feature points between the current laser frame and all the past laser frames(generated during the first mapping)is calculated It is a method to reduce the range of relocation matching.The experiment results show that the relocation method based on laser feature matching has a significant effect on improving the speed of robot relocation.In order to solve the problem of low positioning accuracy caused by the change of environment,a map updating method based on unique grid rate is proposed.Firstly,according to the unique grid rate of the sub map,it is determined whether to replace it with a new submap that can cover it,so as to maintain the high consistency between the built map and the actual environment,and then improve the positioning accuracy of the robot.The experiment shows that the map updating method has a significant effect on improving the accuracy of robot positioning and mapping.In addition,a dynamic pose map method is introduced,that is,the redundant nodes and subgraphs in the global pose map are deleted to reduce the system memory consumption.This paper contains 57 figures and 80 references.