Calculating energy levels for kids' activities
Energy-Meter tool for dance and similar exercises
The Akaike Edge
Experienced ML and DL Ops teams
Efficient Deployment 0%
Ongoing Maintenance 0%
Defining an energy score for the energy meter
Energy estimation is a function of the rate of change of a person's position and gestures from one frame to another. Videos are typically captured at 24/30 FPS and the approach required only a subset of the frames sampled at equal time intervals to compute the energy. The reduction in the frame consumption for energy computation helped in reducing the requirement for heavier hardware.
Selecting the appropriate pose estimator and fetching the landmark points
Our solution used a lightweight pose estimation model to capture the poses for each individual in the frame and compute the energy as a factor of the effort. Various iterations with Media-pipe, mmPose, movenet, Posenet, and NCNN-based pose-estimators were done to improve performance on the video frames, and the estimated landmark points were further fed to the energy-computing function.
Deploying an edge solution with an android application
The deployment was on an edge device which required building an android application that takes video frames as input and renders energy on the screen. Despite the hardware being lightweight and only capable of executing single-pose estimators, we were able to simulate multi-pose estimators for the simultaneous calculation of energies for two users.
Research shows that 87% of Vision AI
projects do not yield expected results
either owing to training data insufficiencies stalling the project, or being too slow to deploy. Our AI experts can help you accelerate in data sparse environments.