PuzzleBoard: A New Camera Calibration Pattern with Position Encoding

TL;DR

PuzzleBoard addresses the limitations of traditional checkerboard calibration by embedding a lightweight, edge-centered position encoding based on two de Bruijn sub-perfect maps into the checkerboard pattern. The four-step decoding pipeline—corner detection, neighbor identification, grid reconstruction, and position decoding—enables robust, error-corrected localization even under occlusion and at low resolutions, while preserving compatibility with existing checkerboard workflows. Experimental results demonstrate multi-target decoding, resilience to downsampling and rotation, and real-time performance that surpasses OpenCV's chessboard detector, highlighting its potential for camera calibration, pose estimation, and marker-based localization in embedded and multi-camera setups. The approach offers high-density reference points and scalable pattern sizes, enabling precise calibration and localization across diverse applications.

Abstract

Accurate camera calibration is a well-known and widely used task in computer vision that has been researched for decades. However, the standard approach based on checkerboard calibration patterns has some drawbacks that limit its applicability. For example, the calibration pattern must be completely visible without any occlusions. Alternative solutions such as ChArUco boards allow partial occlusions, but require a higher camera resolution due to the fine details of the position encoding. We present a new calibration pattern that combines the advantages of checkerboard calibration patterns with a lightweight position coding that can be decoded at very low resolutions. The decoding algorithm includes error correction and is computationally efficient. The whole approach is backward compatible to both checkerboard calibration patterns and several checkerboard calibration algorithms. Furthermore, the method can be used not only for camera calibration but also for camera pose estimation and marker-based object localization tasks.

Figures (8)

• Figure 1: Construction of a sub-perfect map of type $(501,501;\; 3,3)_4$ based on two $(3,3)_2$ de Bruijn rings. Every $3$$\times$$3$ pixel pattern in (d) is unique.
• Figure 2: Starting with a checkerboard (a), circles with diameter $\frac{1}{3}$ of the edge length are placed on the horizontal and vertical edges (b), grouped in blocks of size $167$$\times$$3$ and $3$$\times$$167$ (c), and the bit patterns $\mathrm{A}$ and $\mathrm{B}$ are added (d) to get the PuzzleBoard (e).
• Figure 3: The 18 bits of each $3$$\times$$3$ pattern (a) are unique if the orientation is known. Using all 24 bits of a $3$$\times$$3$ pattern (b) adds rotational uniqueness for $99,33\%$ of the patterns, while $4$$\times$$4$ patterns (c) are always unique under rotation.
• Figure 4: A PuzzleBoard pattern with neighboring PuzzleBoard corners $T_1,T_2$ is projected onto an image by a pinhole camera with camera center $C$. The lines $CB$, $T_2A$ and $T_1P_1$ are parallel to the image plane. The point $D$, whose projection onto the image plane lies at the center between the projection of two neighboring PuzzleBoard corners $T_1$ and $T_2$ divides $\overline{T_1T_2}$ into two segments of length ratio $|T_2D|/|DT_1|=|T_2P_2|/|T_1P_1|=|AP_2|/|T_1P_1|=|CA|/|CT_1|$. If $T_2$ is further away from the camera plane $CB$ than from $T_1$, then $|BT_2|>|T_3T_2|$ for a point $T_3$ with $|T_3T_2|=|T_2T_1|=\frac{1}{2}|T_3T_1|$. Then it follows, that $|CA|/|CT_1|=1-|AT_1|/|CT_1|=1-|T_2T_1|/|BT_1|\geq 1-|T_2T_1|/|T_3T_1|=\frac{1}{2}$. Thus, under the weak assumption that the next corner point $T_3$ is still in front of the camera, the center point of the image of a PuzzleBoard edge is the projection of a point $D$ which lies inside the second third of the edge $T_1T_2$.
• Figure 5: Left: Sample image showing parts of three different PuzzleBoard targets. Center: Detected and successfully decoded corner points (red, green, blue). Right: Detected corners (red, green, blue) and their position in the total pattern of size $501$$\times$$501$ (use zoom in digital paper). The invisible or not detected corner points of the three calibration boards are shown in gray.