基于Kinect深度信息的实时三维重建和滤波算法研究

Research of 3D reconstruction and filtering algorithm based on depth information of Kinect

分析了基于Kinect输出的深度数据进行场景实时三维重建的算法。针对实现过程中出现的深度图像噪声过大的问题,根据其信号结构的特点给出了改进的双边滤波算法。新算法利用已知的深度图像噪声范围,将权值函数修改为二值函数,并结合RGB图像弥补了缺失的深度信息。实验表明,新算法无论是在降噪性能还是计算效率上,都大大优于已有的双边滤波,其中计算速度是原始算法的6倍。

This paper analyzed and improved 3D reconstruction algorithm using the depth information from Kinect.To reduce noise,it proposed an improved bilateral filtering algorithm based on the signal structure.This new algorithm used a two-valued function to compute the weights of the filter,because the range of depth image data was already known.It also combined the RGB values and depth information of surrounding pixels to complement some missing depth information.The results show that the proposed algorithm has much better performance and efficiency,namely 6 times as fast as the original algorithm.