Research On Unmanned Vehicle Combined Positioning System Based On Vision And IMU

Unmanned driving is an important part of the intelligent transportation system,and the precise positioning of vehicles is the core of unmanned driving.Due to the urban canyon effect,tunnel occlusion and other factors,the global positioning system is still unable to meet the high-precision real-time positioning demands of unmanned vehicles.There is also the defect of scale uncertainty when the vehicle pose is obtained by using only monocular visual information.Based on the above-mentioned problems,this paper takes the independent modified unmanned vehicle of Xi’an Technological University as the research object,and conducts in-depth research on the combined positioning system based on vision and IMU.In order to solve the time-consuming problem caused by descriptors matching in the feature matching method,this paper proposes a vehicle positioning algorithm based on adaptive image information correlation fusion.This method combines the optical flow tracking method and the feature matching method by identifying the features loss rate and the rotation matrix parameters.Then a motion pose estimation combined with Random Sample Concensus is performed for the matching results,and the estimated pose is optimized by minimizing the reprojection error.The simulation results show that this method improves the accuracy and efficiency of vehicle pose in visual positioning.Aiming at the problem of scale uncertainty in monocular vision positioning method,this paper proposes a vehicle combined positioning system based on vision and IMU.The system uses a pre-integration strategy to independently integrate the inertial data with fixed image acquisition interval,and fix the inter-frame pose to the previous frame coordinate system.This overcomes the long-term drift caused by the accumulated error in the process of solution in the traditional strap-down inertial navigation system.At the same time,the visual measurement error model and the IMU error propagation model are established by combining with the pose information obtained in the visual positioning.By minimizing the designed objective function,the Gauss Newton method is used to solve the batch estimation problem.It makes up for the shortcomings of the filtering method based on Markov hypothesis and improves the accuracy of the vehicle pose.In order to verify the effectiveness of the combined vision and IMU positioning system,this paper uses an unmanned vehicle experimental platform to make campus road datasets,and conducts multiple sets of trajectory experiments on the datasets and public datasets respectively.The results show that the combined positioning system adopted in this paper is reasonable and effective,which can meet the requirements of various real road motion environments,such as straight line motion,curve motion with a rapidly changing angle of view,large-scale circular roads motion and so on,and realizes the precise positioning of unmanned vehicles.