Home/Work/CARA
CARA logo
CARA
AI & ML / Sports Tech United StatesWebInternal tools

AI computer vision for cheerleading competitions.

An AI system that detects, tracks, and classifies cheerleading actions in competition footage — built for coaches, judges, and analysts.

Timeline
June 2022 – March 2026
Market
United States
Platform
Web & AI backend
Role
Full AI/ML lead
Status
Long-term partner
Long-term partnerCARA
Detection, pose, action
Models
PyTorch & TensorFlow
Framework
YOLO v11
Detection
16 actions detected
Classes

The brief

Cheerleading competitions lacked any objective automated tool to detect and classify athlete actions. Judges relied entirely on manual review with no replayable evidence.

The challenge

Sport-specific actions, occluded performers, fast-moving cameras, and a hard requirement for explainable judge-reviewable outputs at live competitions.

Our solution

A PyTorch temporal action recognition pipeline with a custom annotation dashboard — giving judges frame-level evidence for every classification.

What we built

Six things that make CARA work.

01

16-class action detection

Temporal model classifies 16 distinct cheerleading actions from raw competition video.

02

Person detection & tracking

YOLO v11 and pose estimation track athletes throughout entire competition routines.

03

Frame-level evidence

Every classification links to the exact frames that triggered it — fully auditable.

04

Annotation dashboard

Custom web tool for client team to label training data with a built-in reviewer workflow.

05

Active learning loop

Judge overrides and reviewer corrections feed directly back into the training pipeline.

06

Real-time & post-event

Supports both live competition analysis and post-event performance review.

Tech stack

The tools we picked, and why we kept them simple.

Boring tech where boring is right, sharp tech where it counts.

PyTorchTensorFlowOpenCVPythonYOLO v11SlowFast action recognitionMViT modelLabelStudioOptuna
Results

What changed after launch.

92% action detection
Accuracy
1.4M training frames
Frames
4-year partnership
Duration
References on request
Status

Have a project that needs the same level of care?

Get a complete proposal in your inbox. Less than 3 minutes, no calls needed.

Get Instant Quote Book a discovery call