Photo
PeopleVision

Active Head Tracking

Tracking a person's head with an active camera and zooming in when the person stops. The head detection uses the smoothed silhouette of the foreground object as segmented using background subtraction(Figure 1). To interpret the silhouette, we use a simple human body model consisting of six body parts: head, abdomen, two hands, and two feet. First, we generate a one-dimensional “distance profile” that is the distance of each contour pixel from the contour centroid, following the contour clockwise. This distance profile is parsed into peaks and valleys based on the relative magnitudes of the successive extrema.  The peaks of the distance transform are used to hypothesize candidate locations of the five body parts: the head, two feet, and two hands. Determination of the head among the candidate locations is currently based a number of heuristics based on the relative positions of the candidate locations and the curvatures of the contour at the candidate locations. More specifically, the following objective function is used to decide the location of the head: 

Picture of Head Detection

Figure 1. Head detection steps. (a) The silhouette information (b) Distance profile showing significant peaks and the radii of curvature at the significant peaks.

equation

where equationequationdenote the co-ordinates of the centroid of the body contour and center of the circle fitted to the contour segment associated with  ith peak. equationequationdenote radius and residue of least square fitting of the ith circle. equation,equation , and equationare weights associated with three components of the objective function. In other words, the objective function hypothesizes that smaller, more circular extrema are more likely to be heads. Similarly, the circles that are higher and vertically more aligned with the center of the body are preferred as heads. The details can be found in paper [1]. 

Picture of result of head detection

Figure 2. Output of the head detector on sample foregrounds obtained from the background subtraction. (a)-(d) show successful detection where as (e), (f) illustrate failure modes.

Some demos of active head traking are shown below. All demo videos are in MPEG1 format.

The quadrants are:
Top left: Static camera view, 
Bottom left: The active camera's view, 
Top right: The trace of the person's path in a 
plan view of the room, 
Bottom right: The active camera.

1. Demo1 (video 12.7MB)
Picture of Head tracking 1

2. Demo2 (video 4.5MB)
picture of head track 2

[1] Face Cataloger: Multi-Scale Imaging for Relating Identity to Location
A. Hampapur, S. Pankanti, A.W. Senior, Y-L Tian, L. Brown, and R. Bolle 
IEEE conference on Advanced Video and Signal Based Surveillance, Miami, FL, July 21-22, 2003. PDF

Other Research Areas: