Background
Video cameras are commonly integrated with variety devices for different proposes, thus, facial expression recognition (FER) becomes an important factor in HCI. FER is used in camera devices to capture the moment of smile, to recognising emotion for psychological studies etc.
Conventional web-cams or video cameras only function in the optimal light environment, therefore FER in non-optimal conditions is challenging us.
Xbox Kinect has a depth sensor, it consists an infrared laser projector and an infrared camera. The depth sensor works in any light conditions and it reflects the distance between camera and the object. By analyzing the video captured from depth sensor, we should be able to perform FER in any ambient light conditions. On the other hand, depth sensor from Kinect provides less human-recognisable image/video than RGB camera, thus it protects and ensures that the subject’s privacy is not being intruded.
Problem Statement
Address the difficulty of FER system working in non-optimal light condition and protect people’s anonymity by using Xbox Kinect depth sensor.
Methodologies
Feature extraction
Action Units(AU) are the fundamental actions of individual muscles. Different facial expression requires different combination of Action Units(AU). One approach is dividing facial image into several blocks, corresponded blocks will be detected and extracted for classification.
Classification
Machine learning algorithms will be used for classification. Training and testing data will be applied with 10-fold cross validation before classification.
Evaluation
Comparing the classification results with previous project which uses same video clips but in RGB color.
