@erik_apichai_vikander Hi Erik I would be delighted to share some insights!
I played around with gesture recognition for several months using combinations of color segmentation, background subtraction, and motion detection. I had some success, but the solution was not robust for scenes with bad lighting/similar colors. I moved onto object detection and the end result was a very fast implementation in C++ that I would train with a database of images of the desired hand gesture. The next challenge was that the database needs to contain many images with different lighting, orientations, and environments of the gesture. I designed a novel method to gather these images (I typically have about ~100,000 per gesture) and a parallelized training process. So from the selection of a gesture to a finished, detectable gesture it takes ~2-3 days (on a 40 core mammoth of a machine). Then I packaged it up and now people have an easy to use library for doing gesture recognition with much more on its way!
Hi everybody! I'm part of the team behind gesture.ai--created to make gesture recognition accessible for developers of all skill levels. Give me a shout out for any questions...
Report
@isabellamongalo What makes Gesture.ai better than existing machine vision SDKs?
@jesse_golomb Thanks for the question Jesse. All you need to leverage gesture.ai is a computer and a 2D camera. We have made it just a few lines of code to deploy.
MATRIX Creator
gesture.ai
MATRIX Creator
MATRIX Creator
MATRIX Creator
gesture.ai