AI-powered pose detection, movement analysis, and performance coaching — delivered through a smartphone and the cloud, no professional coach required.
Sports Technology / AI-based Performance Analytics
Microsoft Azure Cloud
YOLO Pose Detection + MediaPipe Landmark Extraction
Flutter — iOS & Android cross-platform application
The Sports Video Analytics platform enables athletes and sports enthusiasts to perform AI-powered self-analysis using mobile-recorded training videos — providing automated pose detection, movement analysis, posture correction, and performance insights without a professional coach.
The goal: bring professional-grade biomechanics analysis to every athlete using only a smartphone camera and cloud AI — at a fraction of traditional system costs.
Athletes and amateur sports players consistently lack access to the tools professionals use — and existing solutions are expensive, hardware-heavy, and studio-dependent.
The project aimed to replace complex hardware with accessible smartphone technology, powered by AI:
Every training video follows a complete pipeline — from mobile upload to AI-analyzed feedback delivered to the athlete's dashboard.
Cross-platform iOS & Android client with rich AI visualization
Lightweight REST API server with full AI model integration
YOLO + MediaPipe pose detection with OpenCV frame processing
Selected for real-time performance at mobile scale:
Used alongside YOLO for detailed body analysis:
6 critical body joints tracked per frame:
The analytics engine calculates four categories of metrics per training session — giving athletes a complete picture of their movement quality.
The scoring engine combined landmark positioning, angle thresholds, movement smoothness, motion consistency, and frame confidence scores into three athlete-facing outputs:
Every uploaded video passes through an eight-step automated pipeline — from mobile upload to push notification delivery on the athlete's device.
Users upload training videos through the Flutter mobile application.
Videos are stored securely in Azure Blob Storage with metadata tagging.
Background processing job is initiated and added to the processing queue.
FFmpeg and OpenCV extract individual frames from the video stream.
YOLO + MediaPipe processes extracted frames for pose and landmark detection.
Movement analytics, joint angles, stability scores, and symmetry metrics are generated.
JSON analytics response is compiled with scores, recommendations, and visualizations.
User is notified on their mobile device when analysis is complete and report is ready.
AI inference on sports video is compute-intensive. The team overcame GPU memory, parallel processing, and latency challenges with three complementary architectural strategies.
Videos processed in chunks instead of entire uploads at once.
Background workers handled inference off the API request cycle.
Processing jobs distributed across multiple worker instances.
Videos compressed on-device before cloud upload — reducing transfer size and upload time.
Only relevant frames selected for AI inference — skipping redundant frames to cut processing time.
Parallel processing pipelines for concurrent frame analysis across worker threads.
AI inference optimized using GPU-enabled Azure VMs for maximum throughput.
Frequently accessed analytics reports cached to avoid redundant computation on repeat views.
Enabled self-training analytics for athletes at all levels — no studio or coach required.
Single Flutter codebase delivered native performance on both iOS and Android.
YOLO-powered inference delivered fast, accurate pose detection on every video frame.
Automated, personalized training suggestions generated without human coach involvement.
Queue-driven, horizontally scalable cloud infrastructure handles growing athlete volumes.
Historical session comparison and progress tracking drove improved athlete engagement.
Real-time live analysis during training
Multi-player tracking in team sports
3D pose estimation
AI-generated personalized training plans
Edge-device inference on-device
Wearable device integration
Voice-based coaching assistant
The Sports Video Analytics for Self-Training platform successfully combined AI, cloud infrastructure, and mobile technologies to deliver an intelligent sports coaching experience accessible to athletes at every level.
Using YOLO-based pose detection, MediaPipe landmark extraction, Flask APIs, Flutter mobile applications, and Azure cloud infrastructure, the system provided scalable and efficient AI-powered performance analytics.
The project demonstrated how modern computer vision and cloud-native AI systems can transform sports training into a highly accessible, data-driven experience — completely eliminating the need for expensive motion capture hardware or professional studio environments.