RGB-D Camera-based 3D Environment Reconstruction Algorithm

Three-dimensional reconstruction technology is a technology that digitally reconstructs3 D scene and is an important support for intelligent aircraft to explore unknown environment.It can provide a realistic scene model for aircraft obstacle avoidance and 3D trajectory planning applications.In this thesis,a 3D scene reconstruction algorithm based on RGB-D camera is proposed for the demand of small unmanned aerial vehicles for the perception and detection of three-dimensional space.The reconstruction algorithm can be divided into five parts: RGB-D data acquisition and processing,visual odometry,pose optimized,loop closure and 3D mapping.For the acquisition and processing of RGB-D data,the thesis established the mathematical model of RGB-D camera,calibrated the camera and registered RGB-D data,and inpainted the depth image data.For the visual odometry,based on the idea of anisotropic diffusion,the thesis improves the feature extraction algorithm to solve the problem of matching between multiple scenes.The thesis designs a binary descriptor with directions to describe the features and the performance was verified by experiments.In addition,based on the geometric constraints between the matching features,the thesis solves the trajectory of the camera attached to the aircraft and the transformation relationship between multiple scenes.For the pose optimized,the thesis designs keyframe strategy and constructs the leastsquares cost function based on the measurement equation of the visual system,and reduce the noise by using the nonlinear graph optimization method.For loop detection,the thesis designs a bag-of-words model based on the binary visual word to optimize the cumulative noise to build a globally consistent trajectory and map.The experiment in this thesis proves its validity.For the three-dimensional mapping,the thesis builds the 3D sparse point cloud map by the processed RGB-D data with the optimized trajectories and pose information.The root mean square error(RMSE)of the relative position between calculated pose and ground truth is used as the evaluation index to verify the accuracy of reconstruction.Based on the reconstructed 3D sparse point cloud,the thesis realize the octomap.