Research On Uav Localization Algorithm Based On RGB-D SLAM

With the development of artificial intelligence,the application of mobile robots in daily life is becoming more and more popular,which makes an increasing number of scholars join in its research.Autonomous navigation,as the core competence of mobile robots,includes positioning,mapping and path planning which can enable mobile robots to move autonomously in the environment.Autonomous navigation technology in known environments has become mature,but in unknown environments,autonomous navigation is still a hot issue that needs to be studied urgently.To solve this problem,SLAM technology is the best solution which is developing vigorously.SLAM(Simultaneous Localization and Mapping),which aims at the localization and map construction to realize the autonomous navigation of robots in the environment by using its own sensors.SLAM does not depend on any known information provided by the external environment and pre-layout scenarios are neither needed,which plays an extremely important role in autonomous navigation in unknown environments.SLAM research based on RGB-D sensor is called RGB-D SLAM,which is a hot research topic in the field of autonomous navigation of robots.In this paper,the key technology of RGB-D SLAM is deeply studied and improved.It is applied to UAV(unmanned aerial vehicle),an important branch of mobile robot,to achieve precise positioning of UAV.The main research work is as follows:(1)Research on camera imaging model and design of adjustable calibration systemCamera is the carrier of visual SLAM to obtain external information.Accurate camera modeling is the most basic and very important step of visual SLAM.To solve the camera calibration problem,firstly,in terms of algorithm,this paper establishes a camera model,which researches the projection relationship between the 3D world and the image.At the same time,using Zhang Zhengyou's calibration method for reference,the camera's internal and external parameters are obtained.By aligning the RGB image with the depth image,the color information in the RGB-D camera can correspond to the depth information one by one,and the more accurate scale information of the pixels can be obtained.Secondly,in terms of physical structure,this paper designs an adjustable calibration system,which can be used in various application scenarios.It is suitable for any orientation of the camera and can calibrate the camera conveniently and quickly at anytime and anywhere.Finally,the calibration experiment results show that the corresponding relationship between environmental information and image information can be accurately obtained,and the accuratecamera parameters can be obtained.(2)Posture Estimation Based on PROSAC+ICPAccording to the image information obtained by the mobile robot in the unknown environment,the pose transformation of the robot can be roughly obtained.Aiming at the problem of low real-time pose estimation,this paper uses SLAM front-end pose estimation algorithm combined with PROSAC + ICP to improve the real-time performance to a certain extent.Firstly,ORB feature points are extracted from each key frame,and the matching relationship is established according to the distance between images.Secondly,this paper uses PROSAC algorithm to eliminate mismatch points.The experimental results show that it is faster and more stable than RANSAC algorithm.Then combined with ICP algorithm,the pose transformation of mobile robot is obtained according to the constraint relationship of three-dimensional point cloud.Finally,in order to reduce the impact of noise and improve the accuracy of motion estimation,this paper also combines the features of RANSAC extraction surface with the features of PROSAC extraction points to make the constraint stronger and get more accurate pose estimation.(3)Posture optimization algorithmThe front-end pose estimation needs to be optimized because of too few constraints and large errors.This paper improves the existing optimization algorithm.Firstly,the pose map is created according to the position and attitude relations is obtained from the front end,and the optimization speed is accelerated by abandoning the calculation of road markers.Then,the K-tree visual dictionary is introduced to establish the loop constraint relationship for the key frames which are not adjacent to the pose map,so as to reduce the global accumulated error through optimization.Then the whole pose map is optimized globally using g2 o tool,and the optimal pose estimation is obtained by fine-tuning the key frame pose.Finally,the algorithm is validated in TUM dataset,and compared with the classical ORB-SLAM verifying the validity of the algorithm.The precise position information of UAV is obtained by running it on the UAV Intel Aero.