Group Activity Recognition via Computing Human Pose Motion History and Collective Map from Video

Chen, Hsing-Yu; Lai, Shang-Hong

doi:10.1007/978-3-030-41299-9_55

Cited by 3 publications

(4 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The GIRN intra+inter form of our proposed method is already capable of outperforming many previous works (i.e. HDTM [7], CERN [12], PC-TDM [30] and PMH [19]), even though it does not incorporate any CNN-based input as they do (e.g. RGB, optical flow, heatmap).…”

Section: Methodsmentioning

confidence: 93%

“…Accuracy HDTM [7] 81.9% CERN [12] 83.3% SSU [13] 90.6% HRN [14] 89.5% Multi-stream CNN [17] 90.5% PC-TDM [30] 87.7% Multimodal Attention [18] 91.7% ARG [15] 92.6% CRM [22] 93.0% PMH [19] 87.7%…”

Section: Methodsmentioning

confidence: 99%

“…Lu et al [18] complement their CNN plus GRU pipeline with pose features extracted using an existing spatio-temporal attention mechanism designed for skeleton information. Chen and Lai [19] also use estimated poses to build a heatmap representation then using an CNN for feature learning, but instead of using a single channel for all joints, each joint has its own channel indicating its potential locations.…”

Section: Group Activity Recognitionmentioning

confidence: 99%

“…Some authors aspired to complement, or even replace, the commonly used image visual features with explicit pose information from the individuals [17,18,19]. However these works do not directly leverage this type of information when reasoning about the individuals interactions, they basically use the pose as an extra/alternative stream for feature extraction.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Skeleton-based Relational Reasoning for Group Activity Analysis

Perez,

Liu,

Kot

2020

Preprint

View full text Add to dashboard Cite

Research on group activity recognition mostly leans on standard two-stream approach (RGB and Optical Flow) as their input features. Few have explored explicit pose information, with none using it directly to reason about the individuals interactions. In this paper, we leverage the skeleton information to learn the interactions between the individuals straight from it. With our proposed method GIRN, multiple relationship types are inferred from independent modules, that describe the relations between the joints pair-by-pair. Additionally to the joints relations, we also experiment with previously unexplored relationship between individuals and relevant objects (e.g. volleyball). The individuals distinct relations are then merged through an attention mechanism, that gives more importance to those more relevant for distinguishing the group activity. We evaluate our method in the Volleyball dataset, obtaining competitive results to the state-of-the-art, even though using a single modality. Therefore demonstrating the potential of skeleton-based approaches for modeling multi-person interactions.

show abstract