基于深度学习的群体行为识别: 综述与展望

Group activity recognition based on deep learning: Overview and outlook

中文关键词  群体行为识别  深度学习  层级时序建模  交互关系推理  Transformer

英文关键词  group activity recognition  deep learning  hierarchical temporal modeling  interaction relationship reasoning  Transformer

中文摘要

      群体行为识别是计算机视觉领域中备受关注的研究方向, 在智能监控系统和体育运动分析等领域中具有广泛的应用推广价值. 本文对过去七年来基于深度学习的群体行为识别方法进行了全面综述, 有助于更好推动群体行为识别的发展. 首先, 介绍群体行为的定义、通用识别流程以及主要的挑战; 其次, 从群体行为识别的建模方法和内在机理进行划分, 并进一步细分类、讨论和分析这些方法的优缺点; 然后, 给出群体行为识别的常用数据集, 列举了相关的开源代码库和评估指标; 最后, 对该领域未来的研究方向进行了展望.

英文摘要

      Group activity recognition has attracted much attention in the computer vision community, and it is widely applied in intelligent monitoring systems and sports video analysis. This paper provides a comprehensive review of the group activity recognition methods based on deep learning over the past seven years, which will help to promote the development of group activity recognition. First, the definition, the general recognition process, and the main challenges of group activity are introduced; Secondly, we classify the group activity recognition methods in modeling and internal mechanism, subdivide them, and further discuss the advantages and disadvantages of these methods; Thirdly, we present the common datasets of group activity recognition, the relevant open-source code libraries, and the evaluation index; Finally, we analyze the future research directions in group activity recognition.