Submission deadline: 31 July 2026, 11:59 PM IST
A
The Challenge

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.

Download Dataset

Deployment

Deployment requirement

Deployment Requirement

Every submission must include a usable way to run or interact with the solution.

Streamlit appGradio appFastAPI endpointDockerized inference app

You do NOT need enterprise-grade MLOps — just a runnable inference/demo interface.

Bonus credit may be given for:

Docker supportClean READMEBatch inferenceSimple loggingConfigurable thresholdsONNX / OpenVINO optimizationBetter usability

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.

CriterionPoints
Problem understanding & scoping15
Technical execution25
Counting logic correctness20
Result quality / robustness15
Deployment / usability15
Communication & ownership10
Total100

Ready to submit your solution?

Submission deadline: 31 July 2026, 11:59 PM IST