Fall Detection Dataset

Annotated image set with human posture labels (falling, walking, sitting), suitable for training vision models for safety or surveillance.

Download dataset

Size

485 images (.jpg, .png) + 485 annotation files (.txt), simple formats, usable directly

Licence

Open Database + open content

Description

‍

This dataset offers a collection of annotated images for the automatic detection of human falls. The images are divided into two folders (train and Val) and each image is associated with a label file indicating the posture (falling, walking, sitting). The annotations were created via the makesense.ai platform with bounding boxes around people.

‍

What is this dataset for?

‍

Train computer vision models to detect falls in surveillance environments.
Develop alert systems for the safety of elderly people at home or in retirement homes.
Experiment with algorithms for the classification and detection of human activity in videos or still images.

‍

Can it be enriched or improved?

‍

Yes, it is possible to complete this dataset with videos, to add new postures (e.g. lying down, standing still), or to enrich the annotations with the degrees of severity of the falls. Since the format is simple (.txt), it is also easy to adapt or convert to COCO or YOLO.

‍

🔎 In summary

Criterion	Evaluation
🧩 Ease of use	⭐⭐⭐⭐⭐ (Very simple to use - image + text formats)
🧼 Need for cleaning	⭐⭐⭐⭐⭐ (Low: annotations already well structured)
🏷️ Annotation richness	⭐⭐✩✩✩ (Basic: only bounding boxes and simple labels)
📜 Commercial license	✅ Yes (Open Database + free content)
👨‍💻 Beginner friendly	✅ Perfect for starting with object detection
🔁 Fine-tuning ready	✅ Yes, in combination with other action recognition datasets
🌍 Cultural diversity	⚠️ Not specified, may be limited in contextual diversity