《计算机应用研究》|Application Research of Computers

基于动态密集条件随机场增量推理计算的多类别视频分割

Incremental multi-class video segmentation based on dynamic dense conditional field inference

免费全文下载 (已被下载 次)  
获取PDF全文
作者 张晓翔,卢先领,周洪钧
机构 1.江南大学 物联网工程学院,江苏 无锡 214122;2.同济大学,上海 200092
统计 摘要被查看 次,已被下载
文章编号 1001-3695(2020)12-054-3781-07
DOI 10.19734/j.issn.1001-3695.2019.07.0312
摘要 针对现有基于条件随机场(CRF)的多类别视频分割计算量随帧数不断增加的问题,提出了一种用于密集(全连接)CRF推断的快速、全动态推理(inference)算法,并有效地推断出了增量式多类别视频分割中动态密集CRF的最大后验概率(MAP)解决方案。与传统的密集CRF处理视频相比,该方法更适合于在线的机器人增量式视频分割的处理计算。实验结果表明,在多类别视频分割应用中,该动态算法明显快于广为人知的标准密集CRF算法,其计算精度与标准密集CRF算法保持不变。几个多类别视频分割测试证实了本算法的推理效率。该算法不仅限于视频分割,还可应用于诸多类似的增量式动态变化CRF模型中MAP推理计算的优化解决方案。
关键词 视频分割; 密集条件随机场; 计算机视觉
基金项目
本文URL http://www.arocmag.com/article/01-2020-12-054.html
英文标题 Incremental multi-class video segmentation based on dynamic dense conditional field inference
作者英文名 Zhang Xiaoxiang, Lu Xianling, Zhou Hongjun
机构英文名 1.School of IoT Engineering,Jiangnan University,Wuxi Jiangsu 214122,China;2.Tongji University,Shanghai 200092,China
英文摘要 For the problem of the cost of calculation in multi-class segmentation based on conditional random field increases with frame numbers, this paper proposed a fast and fully dynamic algorithm for dense(fully connected) conditional random field(CRF) inference. The algorithm efficiently inferred the maximum a posteriori probability(MAP) solution for a dynamically changing dense CRF model that was applied to multi-class video segmentation. Compared with traditional dense CRF for video segmentation, this method is more suitable for incremental(in-line) robotics video segmentation. The experiment results show that the algorithm is significantly faster than the widely known standard dense CRF algorithm in the application of multi-class video segmentation and it can ensure the same accuracy with them. Several multi-class video segmentation tests confirmed the efficiency of inference of the algorithm. It should be noted that the application of this algorithm is not limited to video segmentation, it also can be used to yield similar improvements in many optimization solutions for MAP in dynamically changing models.
英文关键词 video segmentation; dense conditional random field; computer vision
参考文献 查看稿件参考文献
 
收稿日期 2019/7/3
修回日期 2019/9/1
页码 3781-3787
中图分类号 TP242.62
文献标志码 A