Browse categories
Explore
Fiverr Pro
English
$
USD
I will build custom player and ball tracking for your sports analysis detection
About this Gig
The big platforms already own soccer and basketball. If you are building a product, you do not need another one of those. You need tracking that works on YOUR sport and YOUR footage.
That is what I build.
I am a licensed engineer working in computer vision, building detection and tracking pipelines that hold up on real match video: bad angles, moving cameras, crowded frames, occlusion.
WHAT YOU GET
- Player detection with persistent IDs across the clip
- Ball or object tracking through fast motion and occlusion
- Event detection, heatmaps and positional data as structured output
- Clean documented Python plus JSON or CSV your app can consume
STRONG FIT FOR
Combat sports, racquet sports, gymnastics, hockey, regional and niche sports. Anywhere the major platforms only support the big two and leave you with nothing.
HOW I WORK
Send a sample clip first. Camera height, angle and resolution decide what is achievable, and I will tell you where the ceiling is before you order rather than after.
Message me with your sport and a sample clip and I will tell you what your footage can support.
Programming language:
Python
•
Colab
Frameworks:
SimpleCV
•
Keras
•
PyTorch
APIs:
Other
Tools:
Jupyter Notebook
•
OpenCV
•
TensorFlow
•
Stata
•
Colab
Other Data Science & ML Services I Offer
FAQ
My sport is not soccer. Can you still do it?
That is the point of this gig. Niche sports are where I am most useful, because the major platforms only cover the big ones. Send a clip and your rules and I will scope it.
What camera setup do I need?
A fixed elevated wide shot is ideal. Handheld sideline footage works for detection but degrades ID persistence. Send what you have and I will tell you exactly what it supports before you order.
Can you track the ball through occlusion and fast motion?
Yes, using motion-model interpolation across dropout frames. Accuracy depends on frame rate 60fps is meaningfully better than 30fps for fast-moving balls.
Will this work live, or only on recorded footage?
The packages cover recorded footage. Live inference is achievable and I quote it as a custom offer, since it depends on your hardware and latency budget.
Do I own the output?
Yes. Full commercial rights to the code and trained weights on every package.
Can you handle multiple cameras?
Yes, as an extra. Synchronised multi-camera improves ID persistence considerably when players cross or cluster.
