EV-Action

EV‑Action Dataset Overview

Dataset Characteristics

• Multimodal Data: Includes RGB, depth, electromyography (EMG), and two skeletal modalities.
• High‑Quality Skeleton Data: Captured via a motion‑capture system, providing comprehensive motion information such as joints, trajectories, acceleration, with higher precision and sampling rate.
• EMG Modality Introduced: First action dataset to incorporate EMG, offering complementary information to visual and non‑visual modalities.

Data Collection

• Equipment Setup: 8 Vicon infrared cameras, Kinect‑V2 system, and 4 EMG sensors.
• Capture Environment: A 4.6 m × 4.6 m room; cameras cover a 3 m × 3 m detectable area.
• EMG Sensors: Record 16‑bit EMG signals at 1000 Hz, covering the 20‑450 Hz frequency range.

Dataset Applications

• Action Recognition Framework: Proposes an EMG‑based action recognition framework, demonstrating EMG effectiveness in human action analysis.
• Benchmarking: Provides experimental setups and state‑of‑the‑art baselines for each modality, validating the complementary nature of EMG.

Anticipated Impact

• Expected to make significant contributions to interdisciplinary research in human action analysis, computer vision, machine learning, and biomechanics.