WIDK · Sports AI Company, Beijing

AI Motion Recognitionfor Sports Video Analytics

Camera-only, edge-deployed action recognition. No wearables, no cloud dependency — every frame analyzed on site, structured data out.

Partner With Us Explore Products

Products

In Production

100%

On-device Inference

5-step

Full Pipeline

Core Technology

One pipeline, from pixels to data

A five-step, fully in-house pipeline. Each sport gets its own compact vertical model.

Multi-camera Capture

RTSP / ONVIF / GigE input, multi-angle venue coverage.

Edge Inference

NVIDIA Jetson on-device inference. Runs fully offline.

Action Recognition

Sport-specific vertical models: balls, bodies, full-body keypoints.

Analytics

Stats, heatmaps, scoring and highlight generation in real time.

Structured Output

JSON / CSV / API-ready data for any downstream system.

Product Matrix

Built on the same engine

Proven in production in the China market — basketball arenas and martial-arts training.

01LIVE · Edge-deployed

Basketball AI Action Analysis

Real-time recognition of basketball actions from video streams, with per-player stats and automated highlights.

Shooting splits by zone (3PT / mid / FT)
Dribble count, frequency & possession time
Left / right layup detection
Movement heatmap & burst analysis
AI-cut highlight reels

02LIVE · Martial arts

AI Action Scoring Engine

General-purpose motion capture and scoring engine. Deployed for martial arts; dance scenario in progress; extensible to gymnastics, diving, figure skating.

Full-body keypoint pose estimation
Standard-degree & difficulty auto-judging
Real-time camera capture & scoring
Joint angle / velocity / acceleration analysis
Landing stability & routine completeness

03IN DEVELOPMENT

Smart Fitness Testing

Vision-based fitness testing: automatic rep counting, form judging and report generation for multi-person test flows.

Auto counting: jump rope, sit-ups, pull-ups
Jump & running trajectory analysis
Form standard-degree judging
One-click test reports
Multi-person rotation · no wearables

04HW + SW

Edge Computing Platform

Our in-house Jetson edge box: local AI inference that works with no internet, multi-camera input, data never leaves the venue.

On-device inference · fully offline
Multi-camera input (RTSP / ONVIF / GigE)
Local web management console
Structured data output (JSON / CSV)
Compact · low power · flexible mounting

One engine, transferable across sports — camera-only, no wearablesTennisGolfTable TennisGymnasticsFigure SkatingAthletics

Why WIDK

Edge-first by design

Works Fully Offline

All inference runs on the edge device. No cloud dependency, no video upload — data stays in the venue.

Sport-specific Models

Compact vertical models per sport, transferable across disciplines — not one generic model stretched thin.

No Wearables Required

Pure vision pipeline. Optional heart-rate fusion (Polar) for training workloads.

Full-stack, In-house

From edge hardware to models to application layer — one team, no third-party black boxes.

Partnership

Ways to work with us

Open to technology partners, integrators and distributors worldwide.

SDK / API Licensing

Integrate our action-recognition engine into your own product or platform.

Custom Development

Purpose-built solutions on our tech stack — new sports, new scenarios, new workflows.

Full-stack Delivery

Edge hardware + software, delivered as a complete system. OEM-friendly.

Let's talk

Tell us about your scenario — sport, venue type, and what you want to measure. We reply to every inquiry.

kf@tiyundong.com