Real-Time Factory Safety Vision

Name: Real-Time Factory Safety Vision
Start: 2026-08-28T18:30:00.000Z
End: 2026-09-21T18:29:00.000Z
Location: Bhubaneswar and Rourkela

The alarm went off. Which floor? Which corner? Who is still not wearing a helmet?

Registration closed

21 Sept 2026

Team size

3–6 members

Rounds

2 rounds

Mode

In person

Venue

Bhubaneswar and Rourkela

Registration closed

Already registered? Sign in

Role: Computer Vision / AI Engineer
Format: Evaluation only
Prize: ₹60,000
Deadline: Closed 21 Sept 2026
Registered: 485

Overview

Outcomes

Automatic, continuous checking of PPE compliance instead of periodic manual rounds
Early detection of smoke and fire from existing camera infrastructure
Alerts that reach the right personnel with the affected area and event type attached
Measurably reduced incident response time
Evidence that holds up in dynamic, badly-lit, genuinely messy industrial conditions

Overview Factories face constant risk from unsafe worker practices and hazardous incidents such as smoke and fire, yet detection and response are often delayed or inconsistent. Many industrial sites rely on manual monitoring, periodic inspections or standalone alarms that may not quickly identify unsafe gear usage, or provide timely, location-specific evidence of smoke or fire. The result is slower evacuation, increased damage, and higher risk of injury or worse. The need is for an AI-based real-time monitoring system that operates reliably in dynamic industrial environments, supports fast alerting to the right personnel, and provides actionable insight such as the affected area and the detected event type. The Story A smoke alarm goes off in a plant with fourteen sheds. It is a good alarm. It is doing its one job. It says: smoke. It does not say where, it does not say how much, and it does not say that in shed 9 there are four people and two of them are not wearing helmets. Three minutes get spent finding out. Those three minutes are the whole problem.

Problem statement

Detects safety gear compliance per person — helmets, high-visibility vests, protective footwear, gloves where applicable — and identifies non-compliance
Detects smoke presence and fire presence at an early stage
Operates in real time under real industrial conditions: variable lighting, dust, steam, motion blur, partial occlusion, crowded frames, awkward camera angles
Generates alerts carrying location and context — which camera, which zone, what event type, what confidence
Routes those alerts to the right personnel fast enough to change the outcome
Produces actionable insight, not just a bounding box: affected area, event type, and what to do next

For the Grand Finale, your build should include:

A working detection model covering all three classes — safety gear compliance, smoke, fire — with the dataset and training approach documented
A real-time inference pipeline running on video feeds or recorded footage, with frame rate and latency reported honestly
An alerting system that emits location-tagged, context-carrying alerts to defined personnel roles
A performance report including precision, recall and false-positive behaviour, tested on difficult footage — dust, steam, poor light, partial occlusion
A live demo on realistic factory footage showing detection to alert, end to end, with response time measured

The Grand Finale is judged on:

Detection accuracy across safety gear, smoke and fire — 30%
Real-time performance and latency on live or recorded feeds — 25%
Quality of alerting — 20%
Robustness in dusty, dim, crowded, real industrial conditions — 15%
Demo quality — 10%

Rounds

This challenge runs in 2 rounds. You enter once, and the same entry carries through.

1
Mid evaluation
Submissions open
- In person · Bhubaneswar and Rourkela
- Submit by 15 Oct, 11:59 pm IST
- 8 deliverables to submit
- The top 5 move on to Grand Finale
2
Grand Finale
Coming up
- In person · BPUT, Rourkela
- Submission dates to be announced
- 5 deliverables to submit
- The final round. Results are decided here

Deliverables: Mid evaluation

Your idea in one line
One or two sentences a judge can read in ten seconds: what you are building and who it is for.
What will be working by the finale?
The parts you commit to having running at the Grand Finale, most important first.
Tech stack and hardware you plan to use
Languages, frameworks, models, boards and sensors, and anything you still need to get hold of.
A diagram, sketch or mock-up of your idea
One image that shows how it works: an architecture diagram, a wireframe, or a photo of a sketch. Put anything more in your deck.
What have you built or tested so far?
Optional. Anything already running, measured or prototyped, even if it is rough.
Pitch deck (PDF)
Optional. Export your slides as a PDF. Judges read it on the page, next to your answers.
GitHub repository
A link to your code, if you have some yet.
Demo video
A YouTube, Loom or Google Drive link to a short demo, if you have one. Make sure anyone with the link can watch it.

Deliverables: Grand Finale

A working detection model covering all three classes — safety gear compliance, smoke, fire — with the dataset and training approach documented
A real-time inference pipeline running on video feeds or recorded footage, with frame rate and latency reported honestly
An alerting system that emits location-tagged, context-carrying alerts to defined personnel roles
A performance report including precision, recall and false-positive behaviour, tested on difficult footage — dust, steam, poor light, partial occlusion
A live demo on realistic factory footage showing detection to alert, end to end, with response time measured

Evaluation criteria

15% Problem Understanding & Relevance
20% Innovation & Creativity
20% Solution Approach & Technical Depth
25% Prototype / Implementation & Feasibility
10% Impact, Usability & Sustainability
10% Presentation & Demonstration

Questions & answers

No questions yet — be the first to ask.

More challenges to enter

Prizes are offered and awarded by STPI EmTeK, not by Wooble. Amounts shown are as stated by the host and may include the host’s own valuation of non-cash items. Who places, how and when a prize is paid, and any tax or deduction, are between the winner and the host. A challenge may close, change or be withdrawn before results are declared.