Image processing Android developer

Image processing

Ref: Georges Ryssen

Background and objectives
Achievements and activities
Deliverables and results

As a fitness app user, recording oneself doing the exercises from different angles is common, such as from the ground or from a table. However, to enable the posture analysis algorithm to evaluate the movements accurately, a fixed reference view angle is needed. Therefore, the app uses real-time transposition to convert the recorded views to the reference view, ensuring a consistent posture analysis for the user.

Real-time solution using OpenCV homography in Python. Homography isn't something expensive for computers now, it can be computed frame to frame in real time.

  • OpenCV
  • Python
  • Real-time homography

A functional script usable as a function in other scripts or usable directly in the terminal with its documentation.

Red is the distorsed view, green is the warped view, blue is the reference view

Android developer

Background and objectives
Achievements and activities
Deliverables and results

The goal is to create a library to be integrated in order to offer existing Android fitness applications to benefit from the posture analysis and repetition counting solutions developed by Deepmove.

I worked alone on the library, from scratch. I also made the test app which is integrating the SDK.

  • Kotlin
  • Retrofit
  • MLKit (BlazePose)
  • AWS Amplify
  • Frame by frame encoding
  • Maven publication from Gradle

I wrapped all deepmove API into the android library, with its documentation. I also delivered a test app which is using the library.