基于GFU和分层LSTM的组群行为识别研究方法

提出一种以"关键人物"为核心,使用门控融合单元（GFU,Gated Fusion Unit）进行特征融合的组群行为识别框架,旨在解决两个问题：①组群行为信息冗余,重点关注关键人物行为特征,忽略无关人员对组群行为的影响;②组群内部交互行为复杂,使用GFU有效融合以关键人物为核心的交互特征,再通过LSTM时序建模成为表征能力更强的组群特征.最终,通过softmax分类器进行组群行为类别分类.该算法在排球数据集上取得了86.7%的平均识别率.

Group Activity Recognition Based on GFU and Hierarchical LSTM

This paper proposes a group behavior recognition framework with "key persons" as the core and Gated Fusing Unit (GFU)for feature fusion.Its aim is to solve the following two problems :1)Group behavior information is redundant,focusing on key person behavior characteristics,ignoring the influence of unrelated personels on group behavior.2)The internal interaction relationship is complex within group,GFU is used to effectively model interaction feature centered around the key characters and it is temporally evolved into the group characteristics via LSTM processing.Finally,the group behavior category is classified with Softmax.The algorithm achieves an average recognition rate of 86.7% on the volleyball dataset.