Back to Tao Zhao's Research Page

Self-Calibration of a Camera from Video of a Walking Human

Fengjun Lv, Tao Zhao and Ram Nevatia
{flv|taozhao|nevatia} (at) iris (o) usc (o) edu
ICPR 2002 (International Conference on Pattern Recognition), Quebec city, Quebec, Canada, Aug., 2002.


Analysis of human activity from a video camera is simpli- fied by the knowledge of the camera¡¯s intrinsic and extrinsic parameters. We describe a technique to estimate such parameters from image observations without requiring measurements of scene objects. We first develop a general technique for calibration using vanishing points and vanishing line. We then describe a method for estimating the needed points and line by observing the motion of a human in the scene. Experimental results, including error estimates, are presented.

Directed solution (without matrix manipulation) of camera parameters from 3 orthogonal vanishing points.

Approximate solution with the knowledge of the vertical vanishing point and the horizon line.

Compute vanishing points/lines from static scene structures: interactions marked on the image.

Compute vertical vanishing point and horizon line by a walking human: the lines connecting the heads and and the lines connecting the feet will intersect on the horizon (L); the lines connecting the head and the foot will intersect on the vertical vanishing point (V).

Download PDF file of the paper.

This tool will be ready to download soon.