Robust Pedestrian Height Estimation Using Principal Component Analysis and Its Application to Automatic Camera Calibration

Abstract

This paper presents an improved pedestrian height-based camera calibration method using pedestrian 's walking posture. Existing methods have large errors of extrinsic parameters since they use inaccurate height information for homography estimation. To solve the problem, the proposed method consists of three steps: i) pedestrian pose estimation using principal component analysis, ii) leg-crossing detection from the periodicity of walking motion, and iii) calibration based on homography estimated by height information. The proposed method can be applied in surveillance system.