面向Kinect深度图像的导向滤波算法改进  

Improvement of Guided Filtering Algorithm for Kinect Depth Image

作　　者：赵书芳 汤汶[1,2] 万韬阮 朱耀麟[1,4] 武桐[1] Zhao Shufang;Tang wen;Wan Taoruan;Zhu Yaolin;Wu Tong(School of Electronic Information,Xi′an Polytechnic University,Xi′an710000,China;Bournemouth University;Bradford University;Northwestern Polytechnical University)

机构地区：[1]西安工程大学电子信息学院,西安710000 [2]伯恩茅斯大学 [3]布拉德福德大学 [4]西北工业大学

出　　处：《单片机与嵌入式系统应用》2020年第3期46-49,53,共5页Microcontrollers & Embedded Systems

基　　金：陕西省重点研发计划项目(2018SF-351);陕西省教育厅服务地方科学研究计划(18JC012);陕西省科技厅重点研发计划一般项目(2019GY-098);榆林市科技局科创新城项目(2018-2-24)

摘　　要：针对Kinect采集的深度图像普遍存在噪声和空洞的现象而降低人体动作跟踪和识别的效果及精度的问题,本文提出一种面向导向滤波的深度图像滤波算法。导向滤波的引导图采用双阈值的处理方法,通过对深度图像的直方图中灰度值的分布选取合适的阈值;然后对深度图像进行双阈值处理,生成二值化图像;最后将处理过后的深度图像作为引导图,利用导向滤波算法对空洞区域进行深度填充。实验结果表明,经该算法处理后的深度图像保持了导向滤波平滑保边的优点,由于引入阈值图像作为引导图像,在滤波的同时消除了光照阴影等导致的虚假边缘,并且此算法可以兼顾实时性和视觉效果。Aiming at the existence of Kinect depth image noise and cavity,the effect and accuracy of human motion tracking and recognition system are reduced.A depth image filtering algorithm oriented to steering filtering is proposed in this paper.The guided filtering guide map adopts a double threshold processing method to select an appropriate threshold value by distributing the gray value in the histogram of the depth image,then performing double threshold processing on the depth image to generate a binarized image,and finally processing the processed image.The depth image is used as a guide map,and the hole region is deeply filled by the guide filter algorithm.The experimental results show that the depth image processed by the algorithm maintains the advantage of guided filtering smoothing and edge preservation.Since the threshold image is introduced as the guiding image,the false edges caused by illumination shadows are eliminated while filtering,and the proposed guidance is based.The filtered depth image optimization method can balance both real-time and visual effects.

关 键 词：KINECT 深度图像 阈值处理 导向滤波 空洞修复 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]