Description
The BEHAVE dataset is a video surveillance benchmark for group activity and interaction recognition (e.g. walking together, fighting, meeting). It is captured from a fixed outdoor camera and is used as a baseline for HMM-based activity classifiers and group-behaviour models in the vault.
Modality / size
- Modality: fixed-camera RGB video.
- Subjects / scenarios: outdoor scene with multi-person group interactions.
- Labels: per-frame group-activity annotations.
Used by (papers)
- HMM classifier in vault paper achieves ~93.67% on Behave Dataset with a 100-frame window.
Notes
- Public release from the University of Edinburgh.
- Disambiguation: many "behave / behaves / behaviour" hits in claim text are unrelated to this dataset; rely on capitalisation and citation context.