Project Highlights:
- Implemented MediaPipe for data processing
- Chose Kubernetes in сonditions of uneven workload
- Deployed video rendering presenting straightforward pose analysis
Human Pose Estimation Platform
for Athletes
Business Goals:
A sports technology startup, BeONE Sports, is expanding the reach of specialized training to athletes regardless of their location worldwide. Empowering an advanced training methodology with computer vision, they provide data-driven training programs for athletes and coaches. MobiDev covers a full cycle iOS app development, which in particular, includes processing data received from the existing BeONE Sports ML models, video rendering, and in-app purchases.
Product Description:
BeONE Sports is a subscription-based platform for athletes that allows users to create video records with further workout analysis, focusing on specific and isolated actions to enhance the user’s athletic performance. The app estimates the user’s performance, shows accuracy scoring, and gives tips for technique improvements. Along with that, users can compare their performance with the listed experts or their own workout records.
Applied Technologies:
iOS (15.0+, SwiftUI + UIKit), MediaPipe, Python3, Django, Django-Rest-Framework, Docker, Docker Compose, Docker Swarm, AWS EKS, AWS EC2, AWS K8s, AWS S3, AWS SES, AWS Parameters Store, PostgreSQL, Apple developer program, Сelery, Redis, FFmpeg
Implemented MediaPipe for data processing
- Starting with an iOS app for the MVP stage, BeONE sports is expected to also be available for Android and web in the future. MediaPipe allows the deployment of a single codebase across different operating systems, significantly reducing the development timeline and ensuring consistent functionality across platforms.
- Offline video processing is one of BeONe sports key features. MediaPipe effectively optimizes resource utilization for processing high-dimensional data which also ensures smooth performance on resource-constrained devices.
- Another significant achievement with the implementation of MediaPipe is high data processing speed which takes half the time of the video’s duration. Thus users usually get video analysis results within a few seconds, enhancing the app’s overall responsiveness and user experience.
Chose Kubernetes in сonditions of uneven workload
- Kubernetes has proven to be the optimal solution for BeONE Sports, especially with regards to the anticipated high and variable workload from over 10,000 athletes and coaches. Kubernetes gives the advantage of horizontal scalability, enabling us to handle the surge in user demand. As the user base grows, Kubernetes will automatically scale the app by adding more containers to match the increasing workload, ensuring a responsive and high-performing app even during peak usage. Furthermore, Kubernetes intelligently manages container placement and resource allocation, optimizing resource utilization. This not only helps reduce operational costs, it also ensures the best performance.
Deployed video rendering presenting straightforward pose analysis
- After processing data from ML models and evaluating performance accuracy, the app accomplishes video rendering to provide athletes with easy-to-understand analysis. During this stage, BeONE Sports splits a video into segments, adds scores for each attempt and adds a “skeleton” with key points of the user’s body. Users can add an overlay with a “ghost” figure chosen from the library of professional athletes’ videos or the user’s performance history. Along with these features, users get personalized recommendations to enhance their performance.
MobiDev is clearly an impressive organization. Everything from communications, UI/UX, mobile app development and management skills to strategy and understanding the zeitgeist of technology development…MobiDev has it all.
MobiDev has super capable, young and energetic people working to make the best product possible, and they treat the client’s project like it is their own - which is truly rare.
It’s been a great start to a long relationship.
MediaPipe
for data processing
Kubernetes
to handle uneven workload
Video rendering
providing easy-to-understand analysis
