Facial expressions are extracted using deep convolutional neural networks, a deep learning technique used by cutting edge researchers for highest accuracy
Our methods are pose and lighting invariant, capable of detecting micro and subtle expressions, and trained on a high quality and high volume proprietary datasets
Deep learning techniques are also used in building our proprietary deep models, learning user emotions and behavior, across age, gender and geography
We've amassed a proprietary dataset centered around children between the ages 3 and 16, across multiple ethnicities
For greater accuracy of facial expression recognition, the dataset is preclassified and algorithms trained on Facial Action Coding System (FACS)
Our head motion tracker can calculate displacement in 3D, yaw, pitch and roll movements and an overall head movement intensity with superb accuracy
Our facial landmark detection is based on Constrained Local Models (CLM) and is adapted and trained to suit varying face sizes of children
Our deep learning based face detector is built from ground up and is trained to detect faces of children and adults, with varying degrees of tilt
Our robust pre-processing engine counteracts the large variations in illumination and pose by enhancing the quality of input frames
Faces are tracked using a customized implementation of Dr. Zdenek Kalal's award winning Object tracker (TLD 2.0), enhanced and optimized to track multiple faces in realtime with higher accuracy, and long term tracking under partial occlusions