Simuletic/CCTV_ATM_Robbery_Detection_Dataset_Gun_Knife

CCTV ATM Robbery Detection Dataset: Gun & Knife (free sample)

A free sample of 237 synthetic CCTV images of armed robberies at ATMs, with YOLO bounding boxes for offender, victim, gun and knife, for training and testing ATM robbery detection and weapon detection models on security-camera footage. The labels say not only which weapon is present but who holds it and who is threatened. Made by Simuletic, which builds synthetic CCTV datasets for computer vision.

Full dataset and licence: simuletic.com/datasets/cctv-atm-robbery-detection-dataset

	Free sample (this page)	Full dataset
Images	237 PNG, 1,024 × 1,024 px (118 handgun, 119 knife scenes)	3,000, balanced handgun and knife robberies
Classes	offender, victim, gun, knife	offender, victim, gun, knife
Labels	YOLO .txt, one per image	YOLO
Camera view	high-angle ATM security camera, day and night	exterior walls, vestibules, drive-up kiosks and street-corner ATMs; bright daylight to low light
Licence	CC BY-NC 4.0	one-time licence: research $149 / commercial $590 per 1,000 images

What's in the images

Armed robberies at cash machines, seen from a high-angle ATM security camera. In each scene an offender threatens a victim at the machine with a handgun or a knife, often while the victim is using the ATM. About half of the images show a handgun robbery (118) and half a knife robbery (119). They come from 48 scene setups with up to five images each, in daylight and at night. The images have the look of real security cameras: low resolution, sensor noise, JPEG compression artefacts and washed-out colour in low light.

The labels mark people by role, not only the weapon. A plain weapon detector can say that a knife is in the frame; these labels also say who holds it and who is being threatened, so a model can learn to tell a robbery from two people simply standing close together at an ATM.

Classes and boxes

Class	YOLO id	What it marks	Boxes in sample
offender	0	the person making the threat, closest to the weapon	237
victim	1	the person being threatened, usually at the ATM	235
gun	2	handguns and pistols, often held at chest height	122
knife	3	pocket knives, kitchen knives and other blades	118

Every image has one offender box. A gun is boxed in 121 images and a knife in 116.

Folder layout

CCTV_ATM_Robbery_Knife_Gun_Detection_Dataset/
  images/                     # 237 PNG images (handgun_robbery_var10_img1.png, knife_robbery_var1_img1.png, ...)
  labels/                     # one YOLO .txt per image, same file names
  labels.txt                  # class list: 0 offender, 1 victim, 2 gun, 3 knife
  annotated_example.png       # example image with detections drawn
  not_annotated_example.png   # example image without boxes

Each YOLO label line is class_id x_center y_center width height, normalised to 0–1. This is handgun_robbery_var10_img1.txt (victim, offender, gun):

1 0.368764 0.605138 0.289720 0.769153
0 0.743475 0.621440 0.411533 0.737763
2 0.532973 0.364956 0.113959 0.068032

Quick start (Ultralytics YOLO)

Create a data.yaml next to the CCTV_ATM_Robbery_Knife_Gun_Detection_Dataset/ folder:

path: CCTV_ATM_Robbery_Knife_Gun_Detection_Dataset
train: images
val: images   # small sample: use the same images, or split them yourself
names:
  0: offender
  1: victim
  2: gun
  3: knife

Then train:

from ultralytics import YOLO

model = YOLO("yolo11n.pt")
model.train(data="data.yaml", epochs=30, imgsz=640)

Use cases

• Real-time ATM robbery alerts, for example a silent alarm when a weapon is drawn at a cash machine
• Security analytics for bank lobbies, ATM vestibules and street-corner kiosks
• Role-aware threat detection: telling an offender from a victim at close range
• Gun and knife detection on small, partly hidden weapons in CCTV footage
• Pre-training before fine-tuning on your own ATM camera footage

How the images were made

The images are fully synthetic (computer-generated) by Simuletic: no real people, places or incidents are depicted, so there are no privacy or consent issues.

Limitations. Synthetic images do not cover every camera, lens, compression artefact or scene you will meet in deployment. Validate on your own footage, and for production combine synthetic and real data. This sample has 237 images from 48 scene setups; the full dataset has 3,000 images across more ATM locations and lighting conditions.

Full dataset and custom data

The full CCTV ATM Robbery Detection Dataset: Gun & Knife has 3,000 images, YOLO, classes offender, victim, gun, knife. Licence: research $149 / commercial $590 per 1,000 images. Larger sets, other scenes, camera angles or classes: see custom data or write to contact@simuletic.com.

This free sample is also on Kaggle.

More synthetic datasets from Simuletic

• CCTV Weapon Detection Dataset
• CCTV Knife Detection Dataset
• Weapon Detection: Handgun vs Bag of Chips
• All datasets: simuletic.com/datasets · Hugging Face · Kaggle · GitHub

Licence

This free sample is licensed under CC BY-NC 4.0: you may share and adapt it for non-commercial purposes, as long as you credit Simuletic. The full dataset is sold under a separate one-time research or commercial licence.

Citation

@misc{simuletic_atm_2026,
  author       = {{Simuletic}},
  title        = {CCTV ATM Robbery Detection Dataset: Gun & Knife},
  year         = {2026},
  publisher    = {Simuletic},
  howpublished = {\url{https://huggingface.co/datasets/Simuletic/CCTV_ATM_Robbery_Detection_Dataset_Gun_Knife}},
  note         = {Free sample. Full dataset: \url{https://simuletic.com/datasets/cctv-atm-robbery-detection-dataset}}
}

Questions or feedback: contact@simuletic.com