Action Recognition
التعرف على الأفعال
Action recognition classifies what a person is doing across a short video clip — walking, climbing, lifting, falling, fighting. Modern architectures use 3D CNNs (X3D, SlowFast), video transformers (VideoMAE, MViT), or pose-stream models (ST-GCN, PoseConv3D). For FI Tech, action recognition powers slip-trip-fall detection, fight or aggression alerts in worker accommodation…
Definition
Action recognition classifies what a person is doing across a short video clip — walking, climbing, lifting, falling, fighting. Modern architectures use 3D CNNs (X3D, SlowFast), video transformers (VideoMAE, MViT), or pose-stream models (ST-GCN, PoseConv3D). For FI Tech, action recognition powers slip-trip-fall detection, fight or aggression alerts in worker accommodation areas, and unauthorized-task detection (hot work without permit, tank entry without spotter). Best results combine RGB stream with pose stream — the skeleton handles motion, the RGB handles context like fire or confined-space framing.
