Question 1

What violations does VisionAI detect?

Accepted Answer

VisionAI detects 4 types: PPE violations (no safety helmet, no protective gloves) and behavioral risks (mobile phone usage, smoking/lighter usage). All 4 run simultaneously in every processed frame using separate YOLO detection heads.

Question 2

Do I need special cameras or edge hardware?

Accepted Answer

No. VisionAI works with any standard USB or built-in webcam and runs entirely in the browser. For video upload mode, you simply upload an MP4 or AVI file. No IP cameras, no edge devices, no on-premise GPU servers required for demo and pilot use.

Question 3

How accurate is the detection?

Accepted Answer

Helmet detection achieves >82% mAP@50. Glove detection runs at 70–80% mAP@50. Phone and smoke detection are tuned at 65–75% confidence thresholds to balance precision and recall. All models were fine-tuned on SHWD, SH17, and Roboflow PPE datasets.

Question 4

How fast is the real-time detection?

Accepted Answer

VisionAI flags violations within 1–2 seconds of them occurring in the live webcam feed. This sub-2-second latency covers the full pipeline: frame capture, YOLO inference, bounding box rendering, and incident log creation.

Question 5

What is included in the incident log?

Accepted Answer

Each detected violation creates a record containing: violation type (e.g. 'PPE violation: No Helmet'), timestamp, confidence score, severity level (HIGH or MEDIUM), the YOLO bounding box frame, and the camera/session metadata. All incidents can be exported as CSV or PDF reports.

Question 6

Can I process recorded footage instead of live video?

Accepted Answer

Yes. Upload any MP4 or AVI video file and VisionAI processes it frame-by-frame. Every violation found is extracted with its timestamp and added to the incident dashboard — ideal for reviewing past shift footage or auditing compliance retroactively.

Question 7

What YOLO model version do you use?

Accepted Answer

VisionAI uses YOLOv8, fine-tuned on SHWD (Safety Helmet Wearing Dataset), SH17, and Roboflow PPE datasets. The model runs 4 detection heads in parallel, one per violation class group, across every processed frame.

Question 8

How does the dashboard help with compliance reporting?

Accepted Answer

The Safety Report Dashboard aggregates all detected incidents by type, severity, shift, and hour. You can filter by camera, date range, or violation category. KPI cards show total incidents, high-severity count, and overall compliance rate. Reports can be exported as CSV or PDF for audit and regulatory submissions.

AI-Powered Factory Safety Detection

Does Your Factory Have a Blind Spot?

Unreported PPE Violations

Hidden Fire Risks

Phone Distraction

No Incident Paper Trail

What VisionAI Detects

No Helmet

Key Facts

Other Use Cases

From Setup to Incident Log in 3 Steps

Connect Your Input

YOLO Detection Runs

Incidents Are Logged

Two Input Modes

Live Webcam Detection

Video Upload Detection

Every Violation. Logged. Analysed. Reported.

Frequently Asked Questions

Don't Let Another Preventable Incident Go Undetected