Nshisei/multimodal_multiangle_fall_detection_dataset
Multi-View Multimodal Dataset for Fall and Daily Activity Recognition in Wheelchair Users 🧑🦽 Overview This dataset is designed for research in fall detection and daily activity monitoring specifically for wheelchair users. It contains synchronized multi-view, multi-modal data collected from a simulated private room with sensors including thermal cameras, Intel RealSense, and OAK-D. Key Features: Multi-view recordings from four room corners (Position 1–4) High… See the full description on the dataset page: https://huggingface.co/datasets/Nshisei/multimodal_multiangle_fall_detection_dataset.
Multi-View Multimodal Dataset for Fall and Daily Activity Recognition in Wheelchair Users
🧑🦽 Overview
This dataset is designed for research in fall detection and daily activity monitoring specifically for wheelchair users. It contains synchronized multi-view, multi-modal data collected from a simulated private room with sensors including thermal cameras, Intel RealSense, and OAK-D.
Key Features:
- Multi-view recordings from four room corners (Position 1–4)
- High and low camera angles per position
- Modalities: Thermal, RGB, Depth, and IMU sensor data
- User: Wheelchair-dependent individual
- Rich annotations including usual, falling, caution, and emergency labels
🗺️ Room Layout
The room is sized 4.97m × 3.48m with cameras positioned at all four corners. Each corner has a high and low setup.
- Thermal and Realsense cameras are placed in similar top-down positions.
- OAK-D is placed laterally for side-view capture.
🔥 Thermal Image Examples (FLIR Lepton 3.1R - 160×120)
8FPS,
Note: Position numbers (1–4) are indicated in each image.
🎥 RGB and Depth Examples
📋 Annotations
Each .csv file (e.g., corner1_high_annotation.csv) includes file paths and synchronized sensor measurements collected with SensorLogger.
Sample Header
thermal_filename,realsense_depth,oak_depth,Accelerometer_z,...,label,video_idx