An image processing system detects a moving person or other object of
interest. The system generates a thresholded difference image by
processing a video signal or other type of image signal received from a
camera. The difference image is then segmented into regions bounded by
vertical lines passing through the image, and silhouette candidates are
identified in one or more of the regions. Tensor voting is used to
determine saliency values and corresponding tangents for each of the
silhouette candidates, and the resulting values and tangents are used to
detect the object of interest. In an embodiment in which the object of
interest is a moving person, a neck position of the moving person may be
detected by analyzing a sum of x-components of tangents along a
corresponding silhouette. The detected neck position may then be utilized
to determine a head position and a head size for the moving person.