Efficient method for tracking people

摘要

In accordance with one embodiment, a method to track persons includes generating a first and second set of facial coefficient vectors by: (i) providing a first and second image containing a plurality of persons; (ii) locating faces of persons in each image; and (iii) generating a facial coefficient vector for each face by extracting from the images coefficients sufficient to locally identify each face, then tracking the persons within the images, the tracking including comparing the first set of facial coefficient vectors to the second set of facial coefficient vectors to determine for each person in the first image if there is a corresponding person in the second image. Optically the method includes using estimated locations in combination with the vector distance between facial coefficient vectors to track persons.

主权项

1. A method to track persons, comprising the steps of:
(a) generating a first set of facial coefficient vectors by:
(i) providing a first image containing a plurality of persons;(ii) locating faces of persons in said first image; and(iii) generating a facial coefficient vector for each said face by extracting from said first image coefficients sufficient to locally identify each said face, (b) generating a second set of facial coefficient vectors by:
(i) providing a second image containing a plurality of persons;(ii) locating faces of persons in said second image; and(iii) generating a facial coefficient vector for each said face by extracting from said second image coefficients sufficient to locally identify each face, and (c) generating a third set of facial coefficient vectors by:
(i) providing a third image containing a plurality of persons;(ii) locating persons in said third image;(iii) locating faces of persons in said third image; and(iv) generating a facial coefficient vector for each said face by extracting from said third image coefficients sufficient to locally identify each said face, (d) determining predicated locations in said third image of each said person from the location of each said person for which a first facial coefficient vector was extracted with the location of each said person for which second facial coefficient vector was extracted; (e) determining a vector distance between each said second facial coefficient vector and each said third facial coefficient vector; and (f) tracking the persons within said images, said tracking including using said predicated locations in combination with said vector distance between facial coefficient vectors to determine for each face in said second image which is the corresponding face in said third image, wherein the number of coefficients in said facial coefficient vectors sufficient to locally identify each said face are in the range of 5 to 200 coefficients.