Research On Unmanned Vehicle Positioning Method Based On Vision And Inertial Navigation Information Fusion

In recent years,the positioning technology of visual and inertial navigation information fusion has attracted wide attention due to its low cost which has been gradually applied in daily life.The research on the positioning technology o f visual and inertial navigation information fusion can not only make up for the impact of external environment on the vision system through the inertial measurement unit(IMU),but also make up for the cumulative deviation of IMU by using the image information provided by the camera,so that the positioning technology has good accuracy and robustness.At present,the research on the vision and inertial navigation information data fusion scheme mainly includes two methods: nonlinear optimization and filtering.The former has higher accuracy but consumes computer resources,while the latter has higher real-time performance but the accuracy will inevitably decline.In this paper,the vision and inertial data fusion localization technology based on multi-state constraint Kalman filter is studied,aiming to improve the positioning accuracy of the algorithm while taking into care of the real-time performance.The main content and research plan of this paper are as follows:1.The image processing technology based on visual information is studied,and several different feature point extraction strategies are analyzed through experiments.The most suitable feature point extraction scheme is selected,and then the LK optical flow method is used to match or track the feature points.The feature point extraction and matching methods selected in this paper have better quality feature points and take into account real-time performance..2.The positioning technology based on vision and inertial data fusion is studied.An appropriate system state vector is constructed,and the motion model and observation model of the inertial/visual oemometer are derived.A multi-constrained Kalman filter is designed to estimate the motion state.A key frame screening scheme is implemented,and a set of key frame sequences are selected to judge whether the loop detection is constituted.3.Aiming at the problem of cumulative deviation caused by the traditional multi-state constraint Kalman filter algorithm during long-term motion,the loop detection and correction of the system pose are performed based on the bag-of-words model combined with the key frame pose.After that,the Euroc data set is used to conduct experiments to verify loop detection.4.The software and hardware environment needed for this work was built.Based on the data set,several factors that may affect the accuracy of the positioning algorithm were analyzed and verified by simulation experiments.Then,the positioning algorithm was compared with MSCKF algorithm,and the positioning algorithm was evaluated from the perspective of accuracy and real-time performance respectively.Finally,the positioning effect was analyzed by physical testing combined with unmanned vehicle hardware equipment.The results show that the positioning algorithm studied in this paper has excellent performance in both accuracy and real-time performance.