songhaoda/CCTV_Aggressive_Poses_Fight_Detection_Dataset

CCTV Aggressive Poses & Fight Detection Dataset

This dataset contains synthetic images of aggressive behaviors and fighting poses from CCTV angles. It is pre-labeled with YOLOv8 Pose Estimation format (Bounding Box + 17 Keypoints).

The dataset consists of high-quality, synthetic images designed to help train models to detect violence, anomalies, aggression and pre-fight cues in public spaces.

Need more data? This is a curated sample (~100 frames) created by Simuletic. For our full enterprise dataset with diverse environments, multi-person combat scenarios, and evaluation videos, contact Simuletic at https://simuletic.com.

Key Features

Task: Aggressive Pose Estimation & Fight Detection.

Annotations: YOLOv8 Pose Format. Includes Bounding Boxes + 17-point Skeleton Keypoints (COCO format) for precise body language analysis.

Perspective: High-angle CCTV views (often the hardest angle for standard models).

Privacy-First: Fully synthetic data, eliminating GDPR/privacy concerns associated with real surveillance footage.

Dataset Structure

images/: Folder containing high-quality .jpg frames (e.g., aggressive_pose_1.jpg).

labels/: Folder with YOLO Pose .txt files.

Format: class_id x_c y_c w h kx1 ky1 v1 kx2 ky2 v2 … (Standard YOLO Pose).

data.yaml: Configuration file ready for training.

Quick Start (data.yaml)

We have included a data.yaml file in the root. You can start training immediately with Ultralytics YOLOv8:

YAML path: . # Update to your local path train: images val: images # Using same set for sample purposes

Keypoints shape (17 points, x/y/visibility)

Ethics and Limitations

Synthetic Nature: This is fully synthetic data—no real individuals or events are depicted.

Intended Use: For ethical use in public safety research, security automation, and anomaly detection. Do not use for harmful surveillance or profiling.

Bias & Validation: While synthetic data reduces privacy bias, real-world validation is recommended for production systems.

License

This dataset is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0). You are free to share and adapt it, provided you give appropriate credit to Simuletic.

Citation

If you use this dataset in your research or project, please cite:

@dataset{simuletic_fight_detection_2025, author = {Simuletic Team}, title = {Simuletic Synthetic Aggressive Pose & Fight Detection Dataset}, year = {2025}, publisher = {Simuletic}, url = {https://simuletic.com} }

Links

Website & Custom Data: https://simuletic.com Feedback or questions: Reach out via our website or open an issue here.