Research On Indoor Autonomous Positioning Technology Based On Multi-sensor Integration

With the development of urbanization,the indoor space of human activities has became more complex and huge,and the demand for location services has also expanded from the outside to the interior.The rooms' s complex structure,diverse targets and the influence of human factors require that the indoor positioning technology has good universality,which is embodied in:(1)adapting to the unknown environment and having good autonomy;(2)positioning accuracy meets the demand;(3)low cost and easy to popularize.The positioning techniques based on inertial or visual has better autonomy.Among them,the inertial device can estimate the attitude more accurately,but the navigation will diverge due to the accumulated error.The visual sensor can provide rich environmental information but is susceptible to changes in light.Suppose these two navigation methods can be combined,drawing on each other's strength,a more reliable pose estimation can be obtained.Therefore,for the unknown indoor environment,the thesis will take the wheeled mobile robot as the research object,and realize the autonomous positioning of the mobile robot by combining the three kinds of sensor information contained in the nine-axis MIMU and the monocular camera.The main research contents are summarized as follows:1.For the problem of monocular visual odometry can't accurately estimate the pose of the mobile robot due to the scale ambiguity,a monocular visual odometry method with absolute scale recovery is proposed.This method is based on the ORB image features to establish a local map to solve the problem the problem that the traditional two-two frames pose eatimate method relies too much on the reference frame.In order to improve the accuracy of pose estimation,a bundle adjustment is established based on the graphic optimization model of the pose and the map 3D points based on the calculation of the sliding window.Finally,based on the plane model hypothesis and the camera's height prior information,the region of interest in is matched pixel by pixel according to the homography transformation relationship between the keyframes,and the scale factor is solved by establishing a nonlinear optimization to achieve the absolute scale recovery.Experimental results show that this method can stably perform pose estimation and effectively solve the scale uncertainty problem.2.Aiming at the problem that the method of absolute scale recovery can only correct the error of the translation estimated by visual odometry,the attitude error can't be corrected.A method for correcting the attitude error based on integrating the nine-axis MIMU is proposed.First,the error models of gyroscope,accelerometer,and magnetometer included in the MIMU were established.Under quasistatic conditions,for the accelerometer can accurately measure the attitude by sensing the gravitational field,the error self-correction of the accelerometer and gyroscope is realized.Based on the ellipsoid hypothesis,a method based on recursive least squares error correction parameter identification is proposed to achieve the error self-correction of the magnetometer.Then,reference to the Mahony complementary filtering algorithm,a method based on the two-step EKF filtering for solving the attitude is proposed.To reduce the mutual interference between the attitude angles,this method corrected the attitude angles separately.Experimental results show that the self-correction methods are efficient,and the two step EKF filtering algorithm's calculation result is superior to the Mahony complementary filtering.3.By constructing a KF filter,the attitude of the two-step EKF filtering solution is merged with the attitude estimated by the visual odometry.Then,a dead reckoning system is constructed by combinating the merged heading angle information and the translation with the absolute scale recovered.In order to test the effectiveness of the proposed algorithm,a mobile robot experimental platform is built which is based on the ROS system and XQ-mini robot.Using the ROS's toolkit to measure the camera's instinsic parameters,the extrinsic parameters of camera reference to the MIMU and synchronize the camera and the MIMU.Finally a comparative experiment between a single vision system and a combined dead recking system is designed.Experimental results show that the combined system is more stable and reliable than the single visual system,and it'positioning performance is better too.