Google has now open-sourced an Artificial Intelligence for smartphone and it is named as MediaPipe. It is capable of recognizing hand shapes and motions in real-time. The feature was earlier shown is Computer Vision and Pattern Recognition(CVPR) 2019 conference that was held back in June.
The Google AI blog post also mentioned that the ability to recognize the shape and motion of hands can be a vital component in improving and enhancing the user experience across various platforms. MediaPipe is built by google using 3 AI models. The models are named as BlazePalm, Hand Landmark and gesture recognizer models and they have the following functions:
- BlazePalm also is known as the Palm detector is used to analyze a frame and returning an oriented hand bounding box.
- The hand mark model is responsible for returning 3D hand key points from a cropped image region. The best part of this hand-tracking AI is to identify the gestures
- The gesture recognizer is used for classifying already computed key point configurations into a group of gestures.
The researchers have also mentioned that the Artificial Intelligence open-sourced by google is capable enough to recognize signs like:
- Thumbs up
This will be really helpful to a lot of developers to implement gesture recognition technology to their applications. The aim of Google AI is to enhance the functionality and efficiency of Artificial Intelligence. All the interested readers can check out the source code for the AI which is now available on GitHub. The link to the source code is HERE.