Simuletic/CCTV_ATM_Robbery_Detection_Dataset_Gun_Knife

Synthetic ATM Robbery Detection: Multi-Threat Bounding Boxes (CCTV)

Overview

This is an open-source synthetic dataset specifically engineered to train computer vision models in identifying critical retail security threats, active armed robberies, and victim-offender interactions at automated teller machines (ATMs).

Standard security models often struggle to differentiate between a friendly interaction and a close-proximity physical threat. This dataset addresses that gap by providing realistic, high-angle surveillance frames depicting multi-threat weapons (Handguns and Knives) along with pre-labeled roles for the individuals involved (Offenders vs. Victims).

🚀 Need a comprehensive enterprise deployment? This is a verified sample dataset curated by Simuletic. We engineer high-fidelity synthetic datasets to solve edge cases, severe occlusions, and rare security vulnerabilities in AI training. For customized camera angles, lighting distributions, or expanded multi-person scenarios, visit simuletic.com/datasets.

Key Features

• Multi-Threat Distribution: Features a balanced variety of handgun and knife robbery scenarios under distinct lighting parameters.
• Proximity Role Assignment: Programmatically verified spatial annotations that accurately distinguish the active threat agent (Offender) from the targeted civilian (Victim).
• CCTV Degradation Realism: Implements low-resolution surveillance lens aesthetics, digital noise, compressed JPEG artifacts, and desaturated nocturnal lighting profiles to mimic real-world security hardware.
• 100% Privacy-Compliant: Generated entirely using synthetic pipelines. Contains zero real human likenesses or sensitive biometrics, entirely bypassing GDPR, compliance friction, and privacy risks.
• YOLO Optimized: Fully structured and pre-annotated in standard normalized YOLO bounding box (.txt) format, ready for immediate training pipelines.

Dataset Structure & Classes

The dataset repository matches standard object detection conventions:

• images/: High-fidelity synthetic surveillance image files.
• labels/: Normalized .txt files containing target class indices and bounding box anchors.

Class Map

• 0: offender (The individual initiating the threat, positioned closest to the weapon)
• 1: victim (The individual being confronted or withdrawing cash at the terminal)
• 2: gun (Handguns, pistols, or compact firearms held at chest or rib height)
• 3: knife (Blades, pocket knives, or kitchen knives brandished during the encounter)

YAML Configuration

To initialize training with Ultralytics YOLO architectures, your data.yaml layout should be configured as follows:

path: /path/to/dataset
train: images
val: images

nc: 4
names:
  0: offender
  1: victim
  2: gun
  3: knife


Core Use Cases

    Real-Time ATM Anomaly Detection: Power edge AI models inside bank lobbies or exterior kiosks to trigger immediate silent alerts when a weapon is brandished.

    Behavioral Threat Analysis: Train multi-class neural networks to recognize hostile spatial proximity, physical escalation, and robbery postures before an assault occurs.

    Smart City Surveillance: Enhance city-wide monitoring systems to identify vulnerable street-corner transactions and cash machine environments under low light.

Ethics & License

    Synthetic Nature: This data is 100% computer-generated. No real individuals were recorded, staged, monitored, or harmed in the creation of this project.

    License: CC BY 4.0. You are free to share, adapt, and build upon this data for academic research or commercial production pipelines, provided appropriate attribution credit is given to Simuletic.

Citation

If you utilize this dataset in your models, safety evaluations, or publications, please cite the framework as follows:
Plaintext

@dataset{simuletic_atm_robbery_threat_2026,
  author = {Simuletic Team},
  title = {Simuletic Synthetic ATM Robbery Threat Detection Dataset},
  year = {2026},
  url = {[https://simuletic.com](https://simuletic.com)}
}

Feedback & Collaboration

Looking for custom scene layouts, distinct storefront backgrounds, or specific camera HFOV specifications? Reach out directly via simuletic.com or open a ticket in the Discussion tab here on Kaggle!