面向室内场景三维重建的分区联合双边滤波方法

A Partitioned Joint Bilateral Filtering Method for 3D Reconstruction of Indoor Scenes

基于RGB-D传感器的三维重建在高精度制图中应用广泛,然而深度相机采集的图像存在深度值缺失和噪声等问题,影响了三维点云建图的准确度。针对此问题,提出分区联合双边滤波算法。该算法依据深度图像和灰度图像的左右差异分析像素置信度,并对深度图像中缺失像素做插值处理,实现了对图像基于置信度的分区联合双边滤波。采用Middbur标准数据库和Fast Sensor Motion Dataset数据集进行定性分析和定量对比,证明分区联合双边滤波算法有效平滑了噪声。将该算法应用于三维重建,实验结果表明该算法有效修复了深度图像,提高了三维点云建图和位姿估计的精度。

3D reconstruction based on RGB-D sensor is widely used in high-precision mapping.However,the image collected by depth camera has some problems,such as the loss of depth value and the influence of noise,which affect the accuracy of 3D point cloud mapping.To solve these problems,a partitioned joint bilateral filtering algorithm is proposed in this paper.The confidence of pixels is analyzed based on the Left Right Difference(LRD)of depth image and gray image,the missing pixels image are interpolated in depth image,and the partitioned joint bilateral filtering is realized based on confidence.Middbur standard database and Fast Sensor Motion Datase are used for qualitative analysis and is realized comparison to show that the joint bilateral filtering algorithm can smooth the noise effectively.Finally,the proposed algorithm is applied to 3D reconstruction.Experimental results show that the proposed algorithm can effectively repair the depth image and improve the accuracy of 3D point cloud mapping and pose estimation.