2,391 video sequences across 6 human action classes performed by 25 subjects in 4 environmental scenarios. The dataset categorizes movements into walking, jogging, running, boxing, hand waving, and hand clapping to facilitate motion analysis and temporal feature extraction.
Use Cases
- Train spatial-temporal feature extractors to classify human movements using the 6 action labels
- Evaluate model invariance to scale and clothing changes using the s2 and s3 scenario subsets
- Benchmark 3D convolutional neural networks for temporal action localization across the 25 fps video sequences
Strengths
- 6 distinct action categories: walking, jogging, running, boxing, hand waving, and hand clapping
- 2,391 video sequences captured at a frame rate of 25 fps with 160x120 resolution
- 4 experimental scenarios including outdoor (s1), scale variation (s2), clothing variation (s3), and indoor (s4) settings
- Data recorded from 25 unique human subjects to capture interpersonal movement variation