Robust and Accurate Calibration Point Extraction with Multi-scale Chess-Board Feature Detector - Advances in Image and Graphics Technologies - page 153

Robust and Accurate Calibration Point Extraction
with Multi-scale Chess-Board Feature Detector

Yang Liu 1 , Yisong Chen 2 , and Guoping Wang 2,*

1 Shenzhen Graduate School, Peking University
2 Graphics Lab of EECS, Peking University
Beijiing Engineering Technology Research Center of Virtual Simulation and Visualization
{lypku,chenys,wgp}@pku.edu.cn

Abstract . Chess-board grid has been widely used for camera calibration and the
associated feature point extraction algorithm draws much attention. In this paper,
a multi-scale chess-board feature point detector is proposed, along with a
chess-board matching algorithm for a specific marker used in our 3D recon-
struction system. Experiments show that our method is more robust and accurate
compared to commonly used approaches.

Keywords: Multi-scale chess-board corner detection, camera calibration, 3D
reconstruction, calibration point extraction.

1 Introduction

Modern image-based 3D reconstruction systems reconstruct 3D models from a series of
pictures capturing the objects. In general a specific chess-board grid is printed befo-
rehand and put into the scene to help calibrate the cameras.
The chess-board grid we use for camera calibration consists of two parts (As shown
in Fig. 1 ): a blank square area in the middle for placing objects to be reconstructed, and
a chess-board grid region around the blank area. In Fig. 1b the feature points are circled
in green. The four corners of the grid are designed with distinctive style for identifying
global orientation.
To ensure the integrity and accuracy of reconstruction, we have to capture the objects
in as many views as possible and make sure they are well-focused. In most circums-
tances, the chess-board is arbitrarily placed, as shown in Fig. 2 , and it often occurs that
the board is ill-focused. These make it difficult to extract the chess-board vertices
robustly and precisely. Besides, the variation of image size, the location of the board
and the features of the objects increase the difficulty of calibration feature extraction
as well.
In this paper, we propose a multi-scale chess-board feature detector for robust and
precise detection of the calibration markers to meet these challenges. First, we use this
* Corresponding author.