Real-time Glasses Detection

Recent Updates:

• 15 Aug 2019: Add ROI_2 to the measurement regions to improve robustness

Introduction

This is a light-weight glasses detector written in Python for real-time videos. The algorithm is based on the method presented by the paper in Reference with some modifications. Note that the goal is to determine the presence but not the precise location of glasses.

Requirements

• python 3.6

• numpy 1.14

• opencv-python 3.4.0

• dlib 19.7.0

Method

To determine the presence of glasses, the edgeness value (y-direction) of two important regions on the aligned face are computed. Then a indicator is constructed based on these values to do the classification.

The two measurement regions are shown on the schematic below.

What's Next

A threshold is manually chosen in this specific version, which is based on experiment results. The next goal is to develop an algorithm that can choose the threshold automatically in order to enchance robustness.

Welcome to star, fork and try it on your own! :blush:

Reference

Jiang, X., Binkert, M., Achermann, B. et al. Pattern Analysis & Applications (2000) 3: 9. https://doi.org/10.1007/s100440050002