THE OUTCOME

An Enjoyable Interactive Module

We designed Module with Tensorflow.js and anyone with webcam equipped desktop or phone could use this. This module can be used to estimate multiple poses of a single person. In a security level, we made sure no pose data leaves user’s device. A lot of work went into abstracting away the complexities of the model and encapsulating functionality into easy-to-use methods. Atrous Convolution is used to enable the convolution filters in the subsequent layers to have a wider field of view.

DESIGN THINKING

Animation based Approach

Easily configured animation during movements used. Its ideal use case is for when there is only one person centered in an input video.

FRONT END DEVELOPEMENT

Keypoint Detection

All activities are real-time based. Module detects 17 key points to measure the movements.

USER INTERFACE

Acknowledgement of Rapid Movements

User is notified for rapid movements for a longer time through interactive interface. This encourages user to be in line with coach.

BACK END DEVELOPMENT

Image Size Invariant

Module is designed to be image size invariant, which means it can predict pose positions in the same scale as the original image regardless of whether the scale of image changes.

Case Study: Body movement Recognition Module for a popular health&fitness chain of Singapore.

An interactive Real-time Human Pose Estimation with TensorFlow.js

THE OUTCOME

Other case studies