Abstract:
Camera Calibration is a central step in computer vision, machine vision and photogrammetry applications that deduce metric knowledge about an object’s size, shape and location from 2D digital images. Therefore it has become a fundamental topic of research for robotics, navigation systems, 3D scene reconstruction and surveillance systems. And is being extensively studied in these fields for the past decade especially due to the rapidly advancing technology, new techniques are being proposed constantly in the literature. The Camera Calibration process itself estimates the camera parameters such as camera’s focal length, skew, aspect ratio, principal point (the internal parameters) and rotation and translation (the external parameters).
Due to camera calibration’s vast application in many different fields, no generalized solution exists. So many different types of calibration methods have been proposed; all which are specific to the system they are developed for. But mainly they are categorized into object-based calibration (methods that use a calibration object to calibrate the camera) and self-calibration (methods that do not use any particular calibration object). Therefore which camera parameters are determined to calibrate the camera varies from method to method.
The proposed research work presents a novel idea of using any human face as 2D planar calibration object. It uses the observation that adult human faces have a similar shape and size so a generalized model can be used to calibrate a camera. Thus eradicating the need for a special or complicated calibration object, since a human face is readily available in almost all urban scenes. The proposed methodology uses homography to calibrate camera similar to Zhang’s method [2] and uses vanishing points [20] for correct face’s feature point detections. It is a new contribution as very few methods in literature have used any biometrics to calibrate a camera and none have used a human face as proposed in this research work. The proposed framework successfully automates the calibration to a level that little human effort is required. Experiments have been carried out to validate the proposed framework.
