Real-time Face Tracker using Ellipse Fitting and Color Look-up Table in Irregular Illumination

Real-time Face Tracker using Ellipse Fitting and Color Look-up Table in Irregular Illumination

  In this paper, a real-time face tracker for a service robot is introduced. Color information is very useful for detecting human skin color, and makes it possible to reduce the searching area and searching time. We use the Hand S values of the HSI color model to detect human skin color. To cope with illumination change during tracking in realtime, we use a color look-up-table, which is made up in all illumination change. The region grouping process is applied to a color segmented image to find face candidate blobs. At each frame, we apply pattern matching to every face candidate blob using normalized correlation coefficients to verify the presence of a face pattern. Each face candidate blob is fitted by an ellipse, and its major and minor axes are computed. The direction of the major axis determines the planar rotation angle of a face. The length of the minor axis determines the size of the face template. This method makes the face detection fast and detects the 2D rotation angle. To achieve high reliability of face detection, we use light condition compensation and histogram equalization as a preprocess of pattern matching. This real-time face tracker is efficient for human-robot interaction, e.g. face recognition and eye-gaze tracking systems.