Research On Biomimetic Slam Algorithm Based On IMU And Vision Fusion

Simultaneous Localization and Mapping(SLAM)is a technology in which mobile robots realize autonomously localization and environmental map construction through self-carried sensors in an unknown environments.In order to study intelligent navigation methods,researchers have turned their attention to the biological world and carried out research on bionic navigation methods and systems.In the process of visual SLAM,the mobile robot moves too fast,the environment features are missing and the lighting changes,etc.,which leads to the phenomenon such as blurred images or too few extracted image features,thus reducing the positioning accuracy and robustness of the whole system.The Inertial Measurement Unit(IMU)can measure the motion information of the mobile robot,including speed information and position information.In a short period of time,the fast measurement characteristics of IMU can provide accurate motion information and pose constraints for visual images,which can well compensate for the shortcomings of visual sensors.However,in the long running process,the IMU data will have significant cumulative errors,while the visual sensor data is relatively stable and can eliminate the cumulative errors.Therefore,the fusion of vision sensor and IMU information can well estimate and correct the cumulative error of IMU,so that the position recognition is still effective after the blurring of visual images,light changes and missing environmental features.Based on the animal navigation mechanism,this thesis introduces IMU pre-integration theory to assist visual image matching in pure visual SLAM,which is vulnerable to visual image ambiguity,lack of environmental features and illumination changes,etc.,so as to improve the stability of mobile robot SLAM and its adaptability to complex environment.Aiming at the problem of error recognition in the existing position recognition algorithms,the thesis focuses on the image matching method and position recognition method based on the grid cell model,and proposes a bionic positioning method based on the grid cell model.The method includes image matching and bionic positioning.On the one hand it uses the environment representation ability of the activated region of grid cells to realize the recognition of similar scenes,which effectively reduces the error rate of location recognition.On the other hand,the environment information contained in the grid cell nodes is used to realize the accurate positioning of the mobile robot,which improves the accuracy of the location recognition algorithm.On the basis of the bionic localization method based on the grid cell model,a bionic localization method based on the IMU-assisted vision is proposed,furthermore,a place cell model was constructed by using the spatial localization characteristics of the place cell.Combined localization with grid cell of three scales can reduce the number of frames of the image to be searched,solve the problems of false recognition and uncertain recognition when there are too many similar scenes,and realize the accurate representation of spatial information by place cell.The image matching method of IMU-assisted vision is introduced,which can improve the adaptability of bionic positioning method to complex environment and improve the accuracy of position recognition.Finally,the performance of the proposed algorithm is analyzed by comparing the KITTI and St.Luica public datasets with the real environment.The results show that the proposed algorithm can effectively improve the positioning accuracy,real-time performance and adaptability to the environment.