A Fast and Sub-Pixel Detector for Grid-Like Target in Camera Calibration

Abstract

Sub-pixel detection of target points is the performance bottleneck in camera calibration. Traditional algorithms are computational expensive or low precision when we do camera calibration in sport video analysis. In this paper, we propose a new algorithm to detect the grid-like target (i.e. tennis court in TV broadcasting). It has 3 parts: (1) color histogram based interested point classifier making our method faster; (2) sub-pixel refinement by non-linear least squares method improving the accuracy; (3) extended line scan using interested point as the start/end point finding the final line parameters. Results indicate that our detector is faster (<9ms), more accurate and requires less memory than Hough based algorithms if target is grid-like: "straight lines link together".