Android Pose Skeleton
Real-Time On-Device 33-Landmark Human Pose Estimation with CameraX & ML Kit
A native Android application built with Kotlin, Jetpack Compose, CameraX, and Google ML Kit that performs real-time on-device 33-keypoint human pose detection and skeletal overlay rendering.
The Technical Bottleneck
Running real-time human pose tracking on mobile devices requires low-latency camera frame streaming and accurate coordinate transformation between sensor resolution and device display overlays.
Why Existing Baselines Fail
On-device pose estimation provides privacy-preserving motion signals for fitness, rehabilitation, and behavioral analysis without cloud roundtrips.
Core Formulation & Insight
Integrated CameraX `ImageAnalysis` frame pipelines with Google ML Kit Pose Detection in Kotlin/Jetpack Compose, rendering 33 3D/2D body landmarks and anatomical bone connections on a scaled custom canvas overlay.
Architectural Design
Native Android CameraX to ML Kit inference pipeline with custom Jetpack Compose skeletal overlay rendering.
1. CameraX ImageAnalysis Frame Pipeline
Streams non-blocking camera frames with backpressure handling on Android 7.0+ devices.
2. On-Device ML Kit 33-Landmark Inference
Detects 33 full-body skeletal keypoints directly on device.
3. Coordinate Scaling & Compose Canvas Renderer
Maps sensor coordinates to screen space and draws real-time joint landmarks and limb vectors.
Technical Stack & Systems Engineering
Implementation Highlights
- →Real-time 33-landmark skeletal tracking with smooth coordinate scaling across device aspect ratios.
Datasets & Protocol
Datasets & Corpora
- Live On-Device Camera Streams
Evaluation Protocol
Real-time mobile camera validation of 33-point landmark and skeleton link rendering.
Tracked Evaluation Metrics
Delivered a responsive on-device Android pose estimation app rendering 33 body landmarks in real time.
Real-time full-body skeletal topology
Empirical Analysis & Interpretation
Showcases mobile computer vision engineering in Kotlin and Jetpack Compose.
What This System Contributes
- 01Built the CameraX analysis pipeline, ML Kit integration, and custom skeletal overlay renderer.
Boundary Conditions & Future Work
- ·Tracks single-person primary pose per ML Kit Pose Detection specification.