Application Of Gyroscope In 3D Reconstruction Based On Hash Structure

This thesis researches for indoor 3D scene's reconstruction based on the PrimeSense1.09 depth camera,and for the extensive research using the traditional KinectFusion algorithm which exist defects and doing improvement.Research on the use of traditional KinectFusion which use dense volume to store 3D scene information,the restriction of memory capacity makes KinectFusion can only be used for the reconstruction of small objects in the scene,using Hash structure method to solve the problem of memory space constraints for reference.This thesis employs the Hash structure which allowing for real-time access and hidden surface data updating,and do not need to finalize the design or hierarchical grid data structure.Moreover,the data can be efficiently in and out of the Hash table,realize the data exchanging between GPU and CPU,make up for the defect of the space's insufficient,which provides a convenient for the scalability of the storage space in the motion of a sensor.It can make the new scene information continuously joining together to the existing scene model,and finally realize large-scale scene's 3D reconstruction.Research on the Kinect Fusion algorithm which use point-plane iterative closest point(ICP)algorithm in the pose estimation process,but because ICP algorithm is sensitivity to initial value,and easy to fall into local minimum,and there are cumulative error when using ICP algorithm matching and tracking,which lead to the camera cannot be effectively tracking when the camera suddenly move.On the basis of KinectFusion algorithm based on Hash structure,we put forward a method of combining gyroscope to do extended kalman filter(EKF)data fusion.Using a type of LPMS-CU(OEM)gyroscope sensor to get the camera's pose,and obtain the depth camera pose calculated by the ICP algorithm,using EKF data fusion to obtain a more precise pose,and still be able to achieve effective tracking when camera suddenly move,to improve the robustness of the algorithm.Experimental results show that the improved algorithm can be used to solve the problems of Kinect Fusion algorithm in a certain extent,it can be effectively applied to 3D reconstruction of the indoor scenes,and achieved good results.