电子学报 ›› 2020, Vol. 48 ›› Issue (7): 1380-1386.DOI: 10.3969/j.issn.0372-2112.2020.07.018

• 学术论文 • 上一篇    下一篇

FRFCM聚类与深度优化的RGBD场景流计算

张聪炫1,2, 裴刘继1, 陈震1, 黎明1, 江少锋1   

  1. 1. 南昌航空大学无损检测技术教育部重点实验室, 江西南昌 330063;
    2. 中国科学院自动化研究所, 北京 100190
  • 收稿日期:2019-07-01 修回日期:2020-04-17 出版日期:2020-07-25 发布日期:2020-07-25
  • 作者简介:张聪炫 男,1984年7月出生于河南焦作.分别于2007和2014年在南昌航空大学和南京航空航天大学获得学士和博士学位.现为南昌航空大学副教授,硕士生导师,中国科学院自动化研究所博士后.主要研究方向为图像检测与智能识别.E-mail:zcxdsg@163.com;裴刘继 男,1993年7月出生于安徽宣城.现为南昌航空大学测试与光电工程学院硕士研究生.主要研究方向为图像检测与智能识别.E-mail:492888191@qq.com
  • 基金资助:
    国家自然科学基金(No.61866026,No.61772255,No.61866025);江西省优势科技创新团队计划(No.20165BCB19007);江西省科技创新杰出青年人才计划(No.20192BCB23011);航空科学基金(No.2018ZC56008);中国博士后科学基金(No.2019M650894)

RGBD Scene Flow Estimation Based on FRFCM Clustering and Depth Optimization

ZHANG Cong-xuan1,2, PEI Liu-ji1, CHEN Zhen1, LI Ming1, JIANG Shao-feng1   

  1. 1. Key Laboratory of Nondestructive Testing(Ministry of Education), Nanchang Hangkong University, Nanchang, Jiangxi 330063, China;
    2. Institute of Automation, Chinese Academy of Sciences, Beijing 100190, China
  • Received:2019-07-01 Revised:2020-04-17 Online:2020-07-25 Published:2020-07-25

摘要: 针对现有RGBD场景流计算模型在复杂场景、非刚性运动和运动遮挡等情况下易产生场景过度平滑和运动边缘模糊的问题,提出一种基于FRFCM(Fast and Robust Fuzzy C-Means)聚类与深度优化的RGBD场景流计算方法.首先以图像序列连续帧间光流信息为基准,利用FRFCM聚类算法对输入图像进行初始分割,然后根据深度图像的运动边缘信息优化初始分割结果,提取高置信度的运动分层信息.最后设计基于图像分割的RGBD场景流能量函数,采用金字塔变形策略计算精确的场景流结果.分别采用Middlebury和MPI-Sintel数据库所提供的测试图像集对本文方法和现有的RGBD场景流算法进行综合对比分析,实验结果表明本文方法相对于其他方法具有更好的场景流估计精度和鲁棒性,有效改善了场景过度平滑和运动边缘模糊问题.

关键词: RGBD场景流, FRFCM聚类, 深度信息, 过度平滑, 边缘模糊

Abstract: In order to address the issues of scene over-smoothing and motion edge-blurring caused by the existing RGBD scene flow methods under complex scenes,non-rigid movement and motion occlusions,this paper proposes a RGBD scene flow method based on FRFCM (Fast and Robust Fuzzy C-Means) clustering and depth optimization.First,the optical flow information from the consecutive frames is marked as the benchmark and the FRFCM clustering approach is utilized to obtain the initial segmentation of the input image sequences.Second,according to the motion edge information of the depth image,we further optimize the initial segmentation to extract the high-confidence hierarchical motion information.Finally,an energy function of RGBD scene flow based on image segmentation is designed,and the pyramid warping strategy is adopted to compute the scene flow field.We employ the test sets of Middlebury and MPI-Sintel databases to conduct a comparison experiment between the proposed method and the existing RGBD scene flow methods.The experimental results indicate that the proposed method has better accuracy and robustness of scene flow estimation,especially when dealing with the issues of scene over-smoothing and motion edge-blurring.

Key words: RGBD scene flow, FRFCM clustering, depth information, over-smoothing, edge-blurring

中图分类号: