Behaviour & safety

    CCTV PPE Compliance Detection Dataset

    CCTV dataset for PPE compliance detection: 1,000 images of workers on construction sites, in warehouses and on factory floors, with hard hats, safety vests and gloves worn, worn wrong or missing. 9 classes in YOLO and COCO.

    One-time licence · YOLO + COCO labels · instant download

    Sample frame: CCTV PPE Compliance Detection Dataset
    person · hardhat · hardhat_wrong
    Full set
    1,000 images
    Free sample
    200 images
    Labels
    YOLO + COCO
    Classes
    person, hardhat, hardhat_wrong, no_hardhat, vest, vest_wrong, no_vest, gloves, no_gloves
    Video
    Images only
    Licence
    1,000 images · sign in to see prices
    Larger packs
    5,000–20,000 images: $2,400–$8,500, commercial only

    Free sample

    200 labelled images, free to download. The sample has its own open licence; the full dataset is licensed separately.

    Licence

    One-time licence. No subscription.

    Every licence includes the YOLO + COCO labels. Need it on your own camera angles? Order a custom pack.

    Students & research

    Research

    1,000 images, one-time

    Students and non-commercial research. One lab or course.

    • 1,000 images, YOLO + COCO labels
    • Fixed train / val split
    • Dataset card
    • Non-commercial licence

    One organisation

    Commercial

    1,000 images, one-time

    Train and ship models in a commercial product. One organisation.

    • 1,000 images, YOLO + COCO labels
    • Train / val split and evaluation clips where included
    • Dataset card
    • Perpetual commercial licence for trained models

    Larger packs

    5,000–20,000 images

    $2,400–$8,500commercial

    Commercial licence only; the research licence covers 1,000 images. Same labels and split; three sizes, prices in the table below.

    Packs of 5,000, 10,000 and 20,000 images

    Dataset pack prices, one-time, per dataset. Packs above 1,000 images are commercial only.
    ImagesResearchCommercialOrder
    5,000Not offered$2,400Order 5,000
    10,000Not offered$4,500Order 10,000
    20,000Not offered$8,500Order 20,000
    • Commercial licence only: the research licence covers 1,000 images
    • YOLO + COCO labels, fixed train / val split, dataset card
    • Hard negatives in the same share as the 1,000-image set
    • Masks, pose keypoints or VLM captions where the dataset has them
    • Card or invoice; delivered as a download link within five working days

    Larger packs are assembled for your order and delivered as a download link within five working days. Prices in USD, excl. VAT, one-time, per dataset.

    Prices in USD, excl. VAT, per dataset and one-time. Licence terms in the terms.

    About this dataset

    Most PPE detection datasets answer one question: is there a hard hat in the picture? A site camera has to answer a different one: is this worker wearing their PPE correctly? This synthetic dataset shows workers seen from fixed CCTV cameras mounted 3 to 8 m high and looking down, on construction sites (about 55% of images), in warehouses and logistics yards (about 35%) and on factory floors and workshops (about 10%). Lighting covers daylight, overcast, harsh sun, dusk under floodlights, warehouse fluorescent light and rain, and about a quarter of the images are night infrared. Each frame has 1 to 5 workers, many of them small. About 75% of the images contain at least one violation and 25% are fully compliant. Violations include hard hats worn wrong (chin strap loose, worn backwards), held in the hand or left on a surface, vests worn open or tied around the waist, and missing PPE, plus look-alikes: baseball caps, beanies and hoods instead of a hard hat, and orange or yellow clothing without reflective stripes instead of a vest. Nine classes are boxed in YOLO and COCO format: person, hardhat, hardhat_wrong, no_hardhat, vest, vest_wrong, no_vest, gloves and no_gloves. Labels were made automatically, checked by a vision-language model and reviewed visually. Every image is synthetic: no real workers appear. The free sample has 200 images on GitHub, Hugging Face and Kaggle.

    CCTV PPE Compliance Detection sample 2

    What's in the files

    • 1,000 images, train / val split 900 / 100
    • 9 classes: person, hardhat, hardhat_wrong, no_hardhat, vest, vest_wrong, no_vest, gloves, no_gloves
    • About 75% of images with at least one violation, 25% fully compliant
    • Look-alikes: caps, beanies and hoods; orange or yellow clothing without reflective stripes
    • Fixed CCTV views 3–8 m high: construction, warehouse, factory
    • Day, dusk floodlights, fluorescent, rain and about 25% night infrared
    • 100% synthetic: no real workers

    Used for

    • Construction site PPE monitoring
    • Warehouse and logistics safety analytics
    • Factory floor and workshop compliance
    • Hard hat and safety vest alerts on existing CCTV
    • Safety audits and near-miss reporting

    Facts and citation

    CCTV PPE Compliance Detection: the numbers

    1,000 images in YOLO + COCO, classes person, hardhat, hardhat_wrong, no_hardhat, vest, vest_wrong, no_vest, gloves, no_gloves. Per-class box counts are in the dataset card shipped with the download and summarised on the free-sample card.

    How to cite

    If you use this dataset in a paper or report, cite the dataset page. Research licences ask for a citation.

    @misc{simuletic_ppe_compliance_1k,
      title  = {CCTV PPE Compliance Detection Dataset},
      author = {Simuletic},
      year   = {2026},
      url    = {https://simuletic.com/datasets/cctv-ppe-compliance-detection-dataset},
      note   = {Synthetic dataset. Version 1.0.}
    }

    Questions

    CCTV PPE Compliance Detection: common questions

    Which PPE violations are labelled?

    The Simuletic CCTV PPE Compliance Detection Dataset labels hard hats and safety vests as worn, worn wrong or missing: hardhat_wrong covers a loose chin strap, a hard hat worn backwards, held in the hand or left on a surface, vest_wrong covers a vest worn open or tied around the waist, and no_hardhat and no_vest mark workers without them.

    Where are the boxes drawn?

    hardhat_wrong goes on the hat, no_hardhat on the head, vest_wrong on the vest (at the waist when tied), no_vest on the torso, and gloves and no_gloves on the hands, only for workers at least about 20% of the image height.

    Which camera views and conditions are covered?

    Fixed CCTV cameras 3 to 8 m high looking down on construction sites (about 55% of images), warehouses and logistics (about 35%) and factory floors (about 10%), in day, overcast, harsh sun, dusk floodlights, fluorescent light, rain and about 25% night infrared, with 1 to 5 synthetic workers per frame.

    Can I use this dataset commercially?

    Yes, with the commercial licence ($590 for 1,000 images, one-time): one organisation may train models on it and ship those models in products, perpetually. The $149 research licence covers students and non-commercial work only. Packs of 5,000–20,000 images are commercial only, $2,400–$8,500. Reselling or redistributing the dataset itself is not allowed. The free sample is a separate download under the open licence shown on its Kaggle or Hugging Face page.

    Which formats and models does it work with?

    YOLO (.txt per image) and COCO (.json): the YOLO labels load directly in YOLOv5, YOLOv8 and YOLO11, and COCO works with most other detectors, such as RT-DETR and Faster R-CNN. In the full set, images are 1672 × 941, laid out as images/ + labels/ (train, val), annotations/. Results depend on your model, hardware and cameras, so evaluate on a small set of your own footage.

    Ask or order

    A question, a larger pack, or this scenario on your cameras.

    Tell us the camera type, the classes you detect and what your model gets wrong today. We answer within one working day with the pack price, a quote for your own camera views, or a plain answer.

    Send 5 to 20 photos of your empty scenes afterwards and we render this scenario into them.

    Reply within one working day. No mailing list.

    Behaviour & safety

    Related datasets

    Sample frame from the CCTV Shoplifting Detection datasetBehaviour & safety

    CCTV Shoplifting Detection

    Train a shoplifting detection model: 5,000+ synthetic CCTV frames and 100+ videos with YOLO, pose and VLM labels. Free 400-image sample on Kaggle.

    Size
    5,000+ frames + 100+ videos
    Format
    YOLO + VLM
    Sample frame from the CCTV Fall & Incident Detection datasetBehaviour & safety

    CCTV Fall & Incident Detection

    Synthetic fall detection dataset: 1,000 CCTV images of fallen people plus sitting and kneeling look-alikes, labelled person / fallen_person in YOLO and COCO.

    Size
    1,000 images
    Format
    YOLO + COCO
    Sample frame from the CCTV Fight & Aggression Detection datasetBehaviour & safety

    CCTV Fight & Aggression Detection

    Fight detection dataset: 1,000 synthetic CCTV images with person boxes, 17-keypoint poses and fight boxes in YOLO-pose and COCO, plus look-alike negatives.

    Size
    1,000 images
    Format
    YOLO-pose + COCO
    Sample frame from the Pedestrian Attribute Recognition datasetBehaviour & safety

    Pedestrian Attribute Recognition

    Synthetic pedestrian attribute dataset — a PA-100k alternative with captions and structured labels for ViT, CLIP and Re-ID. Free 996-image sample.

    Size
    10,000 images (50K by quote)
    Format
    JSONL (Descriptive + Structured)
    Sample frame from the Driver Drowsiness Detection datasetBehaviour & safety

    Driver Drowsiness Detection

    Synthetic driver drowsiness dataset with 10,000 images and pixel-perfect eye keypoint annotations. 168 open-source samples. Ideal for fatigue detection AI.

    Size
    10,000 images
    Format
    YOLO + Keypoints

    Need this scenario on your cameras?

    Send 5–20 photos of your empty scenes. Custom packs start at $1,500 for 2,500 annotated images on your exact angles.