Research Of 3D Mapping Algorithm Based On Cloud Robotics

The 3D reconstruction is a classic problem in the field of robotics,3D reconstruction based on vision is a typical compute-intensive task,the implementation of traditional method depends entirely on the local robot,which has disadvantages of low precision and slow execution rate.Aiming at the above problems,this thesis researches the 3D scene reconstruction algorithm with the advantage of computing power and resource sharing of the cloud computing,through combining the robot with the cloud computing,we proposed a 3D scene reconstruction framework based on cloud robot.In the process of 3D scene reconstruction based on cloud robot,the key frames are sent to the cloud.In this thesis,the robot embedded system is i.MX6 quad core ARM processor,the ability of data processing is limited,and the conventional feature matching algorithm can't meet the requirements of real-time.In this thesis,the optical flow method for tracking feature points based on FAST,omitting the feature descriptor extraction,feature matching rate will increase by 5-6 times.Although this method has resulted in a decrease in the accuracy of key frame extraction,it can ensure the integrity of the scene information by increasing the number of key frames.In the process of reconstruction,the closed-loop detection is an important part of global optimization.Taking into account the advantages of deep learning in feature extraction,this thesis proposes a closed-loop detection method based on depth learning model,and the experimental test is carried out using the data set and in the actual environment.The experimental results show that the recall rate of the proposed method can achieve more than 38%when the precision is 100%,which is improved by 50%compared with the existing deep learning method.However,the method based on deep learning is time-consuming,in order to shorten the closed-loop detection time,this thesis uses the method of combining image matching and pose estimation,when the precision of this method is 100%the recall rate is 60%.Because of the closed-loop detection process in embedded system execution speed is very slow,the processes of close-loop detection are managed in the AWS cloud computing platform,and using the idea of parallel processing to further improve the rate of closed-loop detection.On the basis of theoretical research,this thesis combines the low cost robot and the cloud computing,and constructs a 3D scene reconstruction system based on the cloud robot.In this thesis,the efficiency of the system is improved by 4-5 times compared with the computing resources of the local robot.The local robot can achieve the efficiency of the proposed system without global optimization,but in this case,the reconstruction accuracy will be greatly reduced.After the robot moves 10 meters away,the local robot will lead to more than 0.12 meter of positioning error,and the positioning error of this system is about 0.04 meter.With the advantage of resource sharing of cloud computing,this system can realize dual robot collaborative reconstruction.The map merging runs in the cloud,improving the efficiency of construction.Finally,this thesis summarizes the research work,and prospects the future research contents and directions.