Marker-Based Extrinsic Calibration Method for Accurate Multi-Camera 3D Reconstruction

TL;DR

This paper tackles the challenge of accurate extrinsic calibration for multi-camera RGB-D systems, essential for precise 3D reconstruction. It introduces an iterative, marker-based calibration method using a cube-shaped marker that exposes three visible faces per view and enforces orthogonality among faces. The pipeline combines data preparation, feature extraction (clustering, regression, reassignment), and a two-stage graph-based calibration (intra-row and inter-row) with robust outlier rejection and alignment via Procrustes, improving overall registration accuracy. The approach yields substantial improvements in 3D reconstructions for the Tech4Diet project and is applicable to broader volumetric capture scenarios, including body scanning and VR visualization, by providing a robust, sensor-agnostic calibration framework.

Abstract

Accurate 3D reconstruction using multi-camera RGB-D systems critically depends on precise extrinsic calibration to achieve proper alignment between captured views. In this paper, we introduce an iterative extrinsic calibration method that leverages the geometric constraints provided by a three-dimensional marker to significantly improve calibration accuracy. Our proposed approach systematically segments and refines marker planes through clustering, regression analysis, and iterative reassignment techniques, ensuring robust geometric correspondence across camera views. We validate our method comprehensively in both controlled environments and practical real-world settings within the Tech4Diet project, aimed at modeling the physical progression of patients undergoing nutritional treatments. Experimental results demonstrate substantial reductions in alignment errors, facilitating accurate and reliable 3D reconstructions.

Figures (10)

• Figure 1: Example of RGB and depth images of the cube used for calibration.
• Figure 2: Overview of the proposed calibration pipeline. The method is structured into three main modules: Data Preparation, Feature Extraction, and Calibration. Each module includes specific processing stages, starting from RGB-D image acquisition and preprocessing, followed by iterative geometric extraction of cube faces, and ending with intra and inter-row calibration steps. The final output is the extrinsic calibration of the multi-camera system.
• Figure 3: Example of two points erroneously assigned to the same plane.
• Figure 4: Comparison of cluster assignments in different phases of the algorithm. (a) Cube model with points assigned in wrong clusters at the beginning of the algorithm. (b) The same cube after iterating multiple times over the Cluster, Regression, and Reassignment phases.
• Figure 5: On the left, the 3D simulation of the booth using Blender. On the right, the designed booth prototype.