Face-Off
Who We Are
What we do
Predict human emotions and gender from video images...
...using
artificial intelligence deep learning.
Background & Motivations
Desire to understand how deep learning can be applied to video
imaging
to predict various human characteristics and
also how accurate
it can be at doing so.
Questions to Answer
What is the best method to measure facial expressions and gender
using video?
How accurate are the predictions of the deep
learning model?
What are the challenges in detecting
micro expressions or increasing the accuracy of reading human
characteristics from video images?
DEMO
The Deployment
The Model
OpenCV takes image from video
Convert image to grayscale
Convert size of image to 48x48 pixels
Run through prediction model h5 files
Return emotion, gender prediction
Overlay prediction display on real-time video feed
The Model
The Model
The Model
The Model
Deep Learning v Machine Learning
Classifier models method: Accuracy = 39%
PCA shows 2,000 features reduction to 300
Quantization of model weights could reduce file size by 75%
Application flexibility from CV to industrial facilities analysis
Questions Answered
What is the best method to measure facial expressions and gender
using video?
Use the Python library openCV in combination
with tensorflow
Questions Answered
How accurate are the predictions of the deep learning model?
63%
Questions Answered
What are the challenges in detecting micro expressions or increasing
the accuracy of reading human characteristics from video images?
Computational processing power
Depth of detail from video imaging
Background lighting
Position of the face