thanhtrung297/cctv-knife-detection-dataset

---
license: cc-by-4.0
---

This is an open-source synthetic dataset for computer vision object detection, focused on people holding knives in public and semi-public environments, viewed from CCTV and surveillance camera perspectives. It is designed to help train and evaluate YOLO, YOLOv8, YOLOWorld, Detectron, and other object detection models for threat recognition, security analytics, and abnormal behavior detection.

Key Features

Classes: person, knife Annotations: YOLO format (bounding boxes, normalized) Image Type: Synthetic, realistic, CCTV-style angles Scenes: Indoor/outdoor, airports, walkways, corridors, public spaces Purpose: Threat detection, surveillance AI, safety analytics, security CV research Size: 114 high-quality annotated images (sample version)

This is a sample dataset created by Simuletic. Larger knife detection sets (3K+ images) and custom scene generation (security, airport, military, intruder, behavior) are available at https://simuletic.com

images/ → .jpg or .png image files
labels/ → YOLO annotation .txt files (same file name as images)
annotations.csv → (optional) structured label overview

class_id center_x center_y width height 0 0.45 0.55 0.20 0.30 # person 1 0.63 0.60 0.15 0.18 # knife

path: /path/to/data train: images val: images names: 0: person 1: knife

Potential use cases:

Knife detection: Identify knives in CCTV/security environments Threat detection: Detect armed individuals in public spaces Surveillance training: Train security camera anomaly models Synthetic data research: Test synthetic-to-real domain transfer

Ethics & Considerations Fully synthetic — no real individuals or incidents depicted Created to support security, safety, and ethical AI research and implementation May not represent full real-world diversity — see our larger dataset for full diversity.

License Creative Commons Attribution 4.0 (CC BY 4.0) You may share, modify, and use commercially, as long as credit to Simuletic is given.

Citation @dataset{simuletic_knife_detection_2025, author = {Simuletic}, title = {Simuletic Synthetic Knife Detection CCTV Dataset}, year = {2025}, url = {https://simuletic.com} }

Related Links Website: https://simuletic.com Weapon Detection Dataset (previous release) https://www.kaggle.com/datasets/simuletic/cctv-weapon-dataset Github & Hugging Face links coming soon

Questions or custom dataset requests? Visit https://simuletic.com or message via Kaggle / Hugging Face.

许可证：CC BY 4.0

本数据集为开源合成计算机视觉目标检测数据集，聚焦公共及半公共环境下持械人员场景，拍摄视角均来自闭路电视（CCTV）与监控摄像头。其设计初衷为辅助训练与评估YOLO、YOLOv8、YOLOWorld、Detectron等目标检测模型，用于威胁识别、安全分析与异常行为检测。

核心特性
类别：人物、刀具
标注格式：YOLO格式（归一化边界框）
图像类型：合成生成，风格写实，采用闭路电视式拍摄视角
场景类型：室内/室外，涵盖机场、步道、走廊、公共空间等
应用场景：威胁检测、监控人工智能、安全分析、计算机视觉安全研究
数据集规模：本示例版本包含114张高质量标注图像

本数据集为Simuletic制作的示例版本。更大规模的刀具检测数据集（含3000+张图像）及定制化场景生成服务（涵盖安防、机场、军事、入侵、行为分析等场景）可通过https://simuletic.com获取。

数据集目录结构：
images/ 目录：存储.jpg或.png格式的图像文件
labels/ 目录：存储YOLO格式标注的.txt文件，与对应图像文件同名
annotations.csv（可选）：结构化标签概览文件

标注格式示例：
类ID 中心X坐标 中心Y坐标 宽度 高度
0 0.45 0.55 0.20 0.30 # 人物
1 0.63 0.60 0.15 0.18 # 刀具

数据集路径配置：
数据集根目录：/path/to/data
训练集路径：images
验证集路径：images
类别映射：0 → 人物，1 → 刀具

潜在应用场景：
- 刀具检测：在闭路电视/安防场景中识别刀具
- 威胁检测：在公共空间内识别持械人员
- 监控模型训练：训练安防摄像头异常检测模型
- 合成数据研究：测试合成数据到真实数据的域迁移能力

伦理与注意事项：
本数据集为完全合成生成，未涉及任何真实人物或真实事件；旨在支持安防、安全及伦理人工智能的研究与落地。
本示例数据集可能无法覆盖全部真实场景多样性，完整多样性数据集可参阅我们的大型数据集版本。

许可证：知识共享署名4.0协议（CC BY 4.0）。您可对数据集进行分享、修改及商业使用，但需注明原作者为Simuletic。

引用格式：
@dataset{simuletic_knife_detection_2025,
author = {Simuletic},
title = {Simuletic Synthetic Knife Detection CCTV Dataset},
year = {2025},
url = {https://simuletic.com}
}

相关链接：
官方网站：https://simuletic.com
往期发布的武器检测数据集：https://www.kaggle.com/datasets/simuletic/cctv-weapon-dataset
GitHub及Hugging Face链接即将上线

如有疑问或定制数据集需求，请访问https://simuletic.com，或通过Kaggle、Hugging Face私信联系。