Video Metrology without the Image-to-Ground Homography

Abstract

The ability to measure metric information from videos is an essential functionality in security software packages that populate databases with human morphometric and gait descriptors. The configuration of such security systems often involves calibrating the cameras in the surveillance network. For this purpose, markers are placed on the ground in the vicinity of each camera location in order to compute a critical image-to-ground homography. This paper shows how the homography computation can be avoided by describing the key components of a system designed for view-independent video metrology. Experiments on measuring the height of subjects and ground distances from different camera views demonstrate the viability of the approach. In addition, it is shown that the calibration model of the proposed method yields more accurate measurements than the frequently used square pixel camera model.