Research and Application of 3D map Modelling for Indoor Environment Based on Siftgpu

With a view to dealing with three dimensional (hereafter referred to simply as 3D) model of indoor environment based on RGB-D data, this study proposes a method of simultaneous localization and mapping of the mobile sensor. By using the color data and depth data collated from the sensor to produce 3D point cloud data of each frame, and this study uses siftGPU to extract and match the feature points of the image. Furthermore, the Random Sample Consensus (hereafter referred to simply as RANSAC) is used to deal with the problem of reducing the accuracy of the pose estimation due to the large error of image feature matching, when the motion is solved by Singular value decomposition (hereafter referred to simply as SVD) which core method is Iterative Closest Point (hereafter referred to simply as ICP) method. Finally, based on the data obtained from the Loop closure with the method of graph optimization to acquire the final position and the high precision point cloud map and motion trajectory are concatenated. The result of this experiment verifies the feasibility and effectiveness of the method.