| Simultaneous Localization and Mapping(SLAM)technology is a core technology to assist robots to locate autonomously in unknown environments.Binocular visual SLAM has received extensive attention due to its advantages of computing depth information of images and small scale drift.The visual-inertial odometry method realizes the accurate pose estimation of the mobile robot during the movement process by effectively fusing the measurement information of the camera and the IMU sensor.Aiming at the problems that the current binocular visual inertial odometry calculation method has low positioning accuracy and degraded tracking performance in environments with large luminosity changes and low textures,it is mainly improved by adding real-time luminosity calibration algorithm and point-line feature fusion algorithm.The main contents and innovations of this paper include the following aspects:(1)Binocular visual inertial odometry based on real-time photometric calibration.Aiming at the problem that the binocular visual inertial odometry method is difficult to effectively track optical flow in scenes with large luminosity changes,a real-time luminosity calibration algorithm is added to the front-end image processing part.Firstly,the principle of camera imaging is introduced,and the photometric parameters that affect the image brightness are analyzed.Then,the photometric parameters are calibrated in real time.Finally,the image brightness is corrected according to the calibrated photometric parameters to improve the stability of KLT optical flow tracking.The experimental analysis of the algorithm before and after the improvement is carried out in the public benchmark dataset.The experimental verification shows that the improved algorithm can improve the robustness of optical flow tracking and has higher positioning accuracy.(2)Binocular visual inertial odometry based on point-line feature fusion.Aiming at the problem that the binocular visual-inertial odometry method is difficult to detect point features in low-texture scenes,which leads to the inability to use visual information for pose estimation,a binocular visual-inertial odometry framework for point-line feature fusion is constructed.In the front-end visual extraction,point and line features are simultaneously extracted,line feature residuals are added in the back-end nonlinear optimization,and a bag-of-words model combining point and line features is constructed in the loop closure detection module.The experimental analysis of the algorithm before and after the improvement is carried out in the public benchmark dataset,and the experimental verification shows that the improved algorithm can improve the positioning accuracy of the binocular visual inertial odometry and improve the loopback recall rate.(3)Overall algorithm experiment of indoor and outdoor multi-scene.Build the mobile robot hardware platform and data processing software platform required for the experiment,jointly calibrate the binocular camera and IMU and verify the validity of the calibration results,and use the mobile robot platform to record multiple scene datasets in the industrial complex,including linear scenes,Ring scene and large surround scene.Comprehensive experiments are carried out on the overall algorithm in multiple indoor and outdoor scene datasets,and comparative analysis is carried out according to the experimental results.The experimental results show that the overall algorithm has higher positioning accuracy and robustness while ensuring real-time performance.The local improvement of the algorithm has a superimposed effect on the overall error correction. |
