Background

Demographic changes due to the aging population demand the development of technologies for independent elderly living in private homes. Besides, researchers and medical practitioners have long sought the ability to continuously and automatically monitor patients beyond the confines of the doctor's office. Thus, computer-vision based in-home diagnostic and monitoring systems get considerable attention in research and development of health technologies in order to set up a proactive and preventive healthcare model.

Human facial image/video can convey information regarding to expression, mental condition, physiological parameters such as heart-rate and respiratory rate, and symptoms of disease. Thus, this PhD project aims to contribute to the development of facial image-based diagnostic and monitoring systems for in-home patient or elderly by employing computer-vision methods.

Aim/vision
Development of facial image-based diagnostic and monitoring systems using computer-vision methods imply a number of challenges to be addressed. Major challenges include setting up an appropriate network of sensors for facial data acquisition, pre-processing of facial image data, selecting the area of interest as a subset of captured data, extraction of appropriate features from the data, and employing effective machine learning methods to automatically detect clinically important factors from the captured data.

In this project, emphasis will be put on face extraction from video using a quality assessment, facial skin color analysis, face alignment and facial feature tracking in video frames, expression and emotion recognition, and finally, developing clinically relevant systems using relevant facial information.

Expected results
The contribution of the project is expected to be an automatic, low-cost, non-intrusive system for facial image-based medical diagnosis and monitoring of patients.

Picture text

An example of heartbeat peak detection in the heartbeat signal obtained from facial video. Red rectangle presents the automatically detected face in a video frame and blue rectangles present region of interest to track for heartbeat footprint in the face.The '+' marks indicate the detected peaks (implies heart pulse) in the signal.