You are invited to attend our next UW CPAMI Graduate Seminar, which will include two presentations.
Amira Ragab will speak on “A Visual-based System for Driver Distraction Detection and Recognition”.
Rodrigo Araujo will speak on “A Semi-supervised Temporal Clustering Method for Facial Emotion Analysis”.
University of Waterloo
Engineering 5
Room E5-5106
Refreshments will be provided
Please see seminar notices below for more information:
A Visual-based System for Driver Distraction Detection and Recognition
Speaker: Amira Ragab
Date: July 31, 2014
Time: 3:00pm – 3:30 pm
Place: E5-5106 Refreshments will be served
Abstract :
Driver distraction and fatigue are considered the main causes of most car accidents today. In this talk, I will present a visual-based non-intrusive system for detecting and recognizing a driver's distraction. This system consists of a hardware component for capturing the driver’s driving sessions on a car simulator, using Kinect camera, combined with a software component for monitoring some visual behaviours that reflect a driver’s level of distraction. In this system, five visual cues are calculated: arm position, eye closure, eye gaze, facial expressions, and orientation. These cues are then fed into a classifier in order to detect and recognize the type of distraction. The use of various cues resulted in a more robust and accurate detection and classification of distraction, than using only one, and experiments on various sequences recorded from different users show very promising results.
A Semi-supervised Temporal Clustering Method for Facial Emotion Analysis
Speaker: Rodrigo Araujo
Date: July 31, 2014
Time: 3:30pm – 4:00 pm
Place: E5-5106 Refreshments will be served
Abstract :
In this talk, we present a semi-supervised temporal clustering method and apply it to the complex problem of facial emotion categorization. The proposed method, which uses a mechanism to add side information based on the semi-supervised kernel k-means framework, is an extension of the temporal clustering algorithm Aligned Cluster Analysis (ACA). We show that simply adding a small amount of constraints, in the form of must-link and cannot-link, improves the overall accuracy of the state-of-the-art method, ACA without adding any extra computational complexity. The results on the non-posed database VAM corpus for three different emotion primitives (valence, dominance, and activation) show improvements compared to the original approach.