Graphics and Media Lab

Overview

The project is devoted to creation of camera based human-computer interface software. While camera can be found in PC systems in increasing frequency, possibility of interaction between human and computer via visual face tracking seems to be very enticing.

We introduce a novel hands free HCI system that enables to navigate mouse cursor through head movements.

Main project goals include:

• HCI simplification for disabled persons;
• Alternative mouse cursor positioning system for laptops;
• Improving of HCI convenience for common users.

Main project tasks include:

• Face detection and tracking;
• Calculation of head orientation;
• Transformation of head orientation information into cursor movement.

Face Tracking

Face is tracked by two decorrelated detectors: Color-Based Face Tracker and Viola-Jones Face Detector. Both algorithms made a good showing by their robustness and speed. They are not exacting to video quality and resolution.

At first system is initialized with Viola-Jones detector. Then both algorithms start to track the face was found at the previous stage. Detectors use different and completely independent sings of face presence in a frame that enables to combine stably their indications, therefore system robustness improves. For example, significant divergence in detectors' indications gives warning of possible system faults.

Color-Based Detector	Viola-Jones Detector

Calculation of Head Orientation

The algorithm of head orientation estimation based on the fact that Color-Based Face Tracker is invariant by head rotation in all planes while Viola-Jones Face Detector is not. As a result their indications are approximately matched when head is in full-face position whereas they differ when head is lightly turned. Using this divergence the approximate head orientation vector is evaluated.

Cursor Moving

Two modes of mouse control are provided in the system: position based (Direct mode) and speed based (Joystick mode). While using a direct mode cursor position is correlated with approximate focus of attention (according to head orientation) whereas joystick mode binds cursor speed to head rotation angle. System enables to combine these two modes according to user's preferences.

Head Orientation Vector	Head Orientation Vector	Head Orientation Vector

The Project Team

Head:
Dr. Vladimit Vezhnevets

Researchers:
Rostislav Shorgin
Alexander Vezhnevets

Contacts

vision@graphics.cs.msu.ru

Related Publications

V. Vezhnevets, R. Shorgin, A. Vezhnevets. "Mouse Control via User's Head Movements". In Proc. Graphicon-2006, Novosibirsk Akademgorodok, Russia, July 2006
.pdf (380kb)
V. Vezhnevets. "Method For Localization Of Human Faces In Color-Based Face Detectors And Trackers". In Proc. Third International Conference on Digital Information Processing And Control In Extreme Situations, pp. 51-56, Minsk, Belarus, May 2002.
.pdf (83kb) .ps.gz (123kb)