The Research Of Indoor 3D Scene Construction Based On The Kinect

At present,the SLAM(Simultaneous Localization and Mapping)technology is widely applied in various industries and fields.There are more requirements for the efficiency and accuracy of SLAM technology.Although visual SLAM has been under development for the past few years,there are still some problems in local data association,global data association and uncertain information processing.Based on this situation,this paper uses Kinect as a visual sensor to design a fast and high-precision SLAM method,on the basis of acquired image information.we achieve the construction of indoor three-dimensional scene map.The main contents of this paper are as follows:Firstly,we utilize the SURF feature point matching to estimate the camera poses.In this thesis,we limit the total number of extracted SURF feature points to improve the timeliness of the algorithm.At the same time,we use the quadratic tree algorithm to optimize the distribution of feature points in the image in order to improve the robustness of the algorithm.After the initial matching of the feature points.And on the basis of the commonly used distance filtering algorithm,we design a efficient slope model.Then,we use RANSAC method to further filter the feature point matching results,which improves the correctness of feature point matching.Finally,the Gauss-Newton algorithm is used to minimize the projection error to estimate the pose of the camera.Secondly,we utilize the visual dictionary method to conduct the loop closure detection.On the basis of the traditional visual dictionary,the paper introduces the processing of outliers to improve the description ability of visual dictionary.Next,we use the k-means++ algorithm to do the cluster operation so that it can eliminate the local optimization problem of k-means algorithm.Meanwhile we use the principle of triangular inequality to simplify the clustering process.Next,we use the visual dictionary to generate the visual vocabulary vector of the image,and calculate the similarity to find out the historical frame with higher similarity.Then by calculating the number of internal points,we can determine the possibility of closed loop.Finally,we use the graph optimization method to achieve indoor three-dimensional scene construction.In order to reduce the influence of uncertain information,the SLAM problem is transformed into a graph optimization problem by formulating the SLAM problems as a graph model.We utilize Levenberg-Marquardt algorithm and take advantages of the sparse feature of SLAM to complete the optimization work of the graph.On the basis of optimization,we conduct the point clouds registration,then achieve the construction of indoor three-dimensional scene.Through theoretical analysis and experimental verification,we solve the problem of accuracy and computational efficiency of SLAM algorithm to some extent,and the thesis can provide a useful reference or help for the development of visual SLAM technology.