Research On Visual Location Algorithm Based On Depth Camera And Inertial Measurement Unit

With the development of science and technology,people's life style has changed.There are a large number of application requirements for the positioning in the indoor environment.However,GPS-based positioning has been unable to be widely used in indoor environment.Instead,it is the visual positioning method.With the development and progress of visual sensors,depth camera has become a commonly used visual sensor in visual positioning.At the same time,the advantage that depth camera can directly obtain depth information is favored by the majority of researchers.However,the pure vision positioning algorithm is easily affected by the environmental state,resulting in low positioning accuracy.Aiming at the problem of low positioning accuracy of the visual positioning system,the paper adopts the method of depth camera and IMU(inertial measurement unit)fusion to study the positioning algorithm.And the paper improves the algorithm of keyframe selection in the back end of the system,so as to improve the positioning accuracy of the system and reduce the running time of the system.The main research work can be summarized as follows:1.The basic theories of vision inertial positioning system are studied,including the theory of rigid body change in 3D space,the principle and structure of depth camera,camera calibration experiment and nonlinear optimization.2.The motion estimation of visual odometer and IMU are studied.The ORB algorithm is used to extract feature points,and the Brute-Force Matcher is used to match the image.Then RANSAC is used to remove the mismatches.The pose of camera is solved by P3 P method,and the estimated value of pose is optimized by BA.The motion estimation of IMU is deduced,and then the internal parameters of IMU and the external parameters of camera and IMU are calibrated by imu?utils and Kalibr respectively.3.The back end optimization of vision inertial system is studied.The initial drift,velocity and acceleration of IMU are calculated.The keyframe selection method is improved to improve the positioning accuracy and reduce the running time of system.The fusion optimization problem based on the system adopts the sliding window optimization method,and updates the data in the sliding window through the edge method.Finally,the experimental results show that compared with the common algorithm,the improved key frame selection algorithm reduces the positioning error by 31.2% and the total running time by 5S.So,the improved algorithm has higher positioning accuracy and shorter running time.4.The effectiveness and performance of the algorithm are verified in the datasets and the real scenes.The datasets experimental results show that the positioning accuracy of our system is higher than okvis system,and the positioning error is reduced by 15.7%;the total running time is reduced by 6.6s;the real-time performance of the back-end is lower than okvis,and the average time consumption of the back-end is increased by 6.7ms.The experimental results in the real scenes show that the system is effective,real-time performance meets the actual application,and the positioning accuracy is higher than that of the pure vision positioning algorithm.