Our data platform correlates sensor data with scientifically validated movement patterns — the core of our intellectual property.
Two Inteli sensor modules per athlete capture position, acceleration and time. Complemented by BLE wearables and markerless camera motion analysis.
Transfer via Bluetooth Low Energy to the smartphone. The app processes data on-device and transmits as soon as the athlete moves.
Data aggregation, staging and store: structured, semi- and unstructured data feed our data model and matching algorithm.
Results at a glance — via mobile app or web interface, leveled for athlete, coach and medical staff.
Not a concept paper: these building blocks are implemented and in use in our sister projects — from computer vision to mobile to sensor infrastructure.
Markerless motion and pose analysis in real time — on the server, in the browser and directly on the smartphone.
Native Android apps with on-device inference, 60 fps live overlays and offline capability — the smartphone as measuring instrument.
IMU wearables and BLE smartwatch integration, room bridges based on ESP32 and contactless WiFi sensing for fall detection — proven in patient monitoring.
Offline-first data platform with REST APIs and role-based web dashboards — from a gym without internet to the cloud.
InjuryDetector builds on technology developed and tested in three real projects:
Cooperative multi-agent pipeline for live game analysis: real-time pose biomechanics, shot detection, rule monitoring — server-side (FastAPI + YOLO) and 100% locally in the browser (ONNX).
Android app turning a standard smartphone into a 3D motion sensor: OpenCV camera calibration (solvePnP), ML Kit pose tracking and velocity-based jump detection at 60 fps.
Offline-capable monitoring with BLE smartwatches, ESP32 room bridges and contactless WiFi sensing for fall detection — the technological basis for our "Sensors for Seniors" use case.
We track the scientific literature continuously and map it against our roadmap — here are the most relevant developments of the last few months.
Open-source systems such as OpenCap now reach clinically usable kinematic accuracy from smartphone video and have been used to validate ACL re-injury risk factors — confirming our approach of combining IMU data with camera-based motion analysis.
A recent meta-analysis reports a pooled accuracy of 0.92 for wearable-sensor-based machine learning models predicting return-to-sport readiness after lower-extremity injury — a strong external signal for the clinical value of our analysis approach.
New deep learning models estimate Achilles tendon and patellofemoral joint forces from consumer wearable data and confirm that metrics such as ground reaction force and tibial shock are relevant predictors of overuse injury — exactly the kind of metrics our risk index is built on.
Low-cost textile capacitive sensors embedded in insoles, combined with deep learning, achieve over 99% accuracy in gait-phase recognition — a possible future addition alongside our IMU foot sensor modules.
Each of these metrics is documented in the scientific literature as sensitive to age-, load- and injury-related change. InjuryDetector captures them continuously and compares them against the personal baseline.
Gait speed, cadence, step and stride length, step time, double-support ratio, step width — the best-established base KPIs.
Left-right asymmetry and step-time variability react sensitively to injuries and fatigue — often before pain appears.
Turning metrics, weight transfer and the risk index over time make recovery and creeping overload visible.
The starting point was a medical study with an ice hockey team: concussions were examined via MRI scans and movement analysis. Together with leading scientists we identified patterns — such as increased swaying and a wider distance between the feet — that indicate the severity of a concussion.
Correlation of multiple independent variables: sensor data is matched against key indicators and the knowledge database to detect and predict injuries.
Every insight flows back into the database. With every user the algorithm gets more precise — the basis for prevention instead of mere reaction.
Movement phases are modelled as state machines — from jump apex to gait cycle. This makes detections explainable, testable and robust.
Data minimisation as a core principle: processing on-device or on local servers wherever possible, encrypted transfer, no sharing with third parties.
The athlete decides who sees which data — from the personal dashboard to leveled sharing with coaches and medical staff.