Vehicle Counting and Flow Analysis Challenge
Build a computer-vision system that analyzes a fixed-camera video and counts cars crossing a defined line.
Using the provided video, build a working solution that detects cars, counts unique crossings against a defined line rather than counting cars in every frame, and produces four outputs — a total count, an annotated output video, alerts, and logs — plus a simple deployable interface.
Evaluation
What we're scoring you on.
- Practical problem solving
- Computer vision engineering
- Counting / tracking logic
- Ability to ship
- Clarity of communication
- Lightweight production readiness
Build
What you need to build
- ▪Detect cars in the video
- ▪Track or count them consistently
- ▪Define a single crossing line
- ▪Count unique line crossings
- ▪Produce the required outputs: count, annotated video, alerts, and logs
- ▪Package the solution in a simple app/API
Required deliverables
1. Car detection and tracking
detect and track cars across frames of the video
2. Car count
total number of cars that cross the defined line
3. Annotated output video
output video with bounding boxes drawn around detected cars
4. Alerts
an alert raised each time a car crosses the line
5. Logs
a log recording each counted crossing with a timestamp
6. A deployable interface
See the Deployment Requirement below — every submission needs one.
Dataset
Download the challenge dataset
The challenge dataset is a single curated surveillance-style traffic video prepared for this challenge, based on public dataset sources such as VIRAT.
Dataset source
Reference dataset: VIRAT Video Dataset
AetherEdge provides a curated challenge video — you do not need to download the full VIRAT dataset yourself.
What’s inside the dataset
- 1 fixed-camera traffic video
- Line definition guidance
- Expected output examples
vehicle-counting-challenge.zip
For this challenge only. Do not redistribute.
Deployment
Deployment requirement
Deployment Requirement
Every submission must include a usable way to run or interact with the solution.
You do NOT need enterprise-grade MLOps — just a runnable inference/demo interface.
Bonus credit may be given for:
Expected outputs
What a good submission includes
- Total car count
- Annotated output video with bounding boxes around cars
- Alerts for each line crossing
- Log of crossings with timestamps
- Clear description of counting logic
- Demo interface screenshot or video
Rules
Rules for this challenge
- Individual submission only.
- Open-source libraries are allowed.
- Pre-trained models are allowed.
- AI coding tools are allowed, but students must understand the final solution.
- Final submission must include code, demo, and report.
- Final submission must include a deployable interface.
- No face recognition.
- No identity matching.
- No manual hardcoding of final answers.
- Do not redistribute the challenge dataset.
- Late submissions may not be considered.
Rubric
How this challenge is scored (100 points)
Counting logic correctness means showing how you avoid or reduce double-counting, and clearly defining what counts as a line crossing.
| Criterion | Points |
|---|---|
| Problem understanding & scoping | 15 |
| Technical execution | 25 |
| Counting logic correctness | 20 |
| Result quality / robustness | 15 |
| Deployment / usability | 15 |
| Communication & ownership | 10 |
| Total | 100 |