Abstract
Three-dimensional (3D) technologies have been developing rapidly in recent years, influencing industrial, medical, and cultural fields. This paper introduces an automatic 3D human head scanning-printing system that provides a complete pipeline to scan, reconstruct, select, and print physical 3D human heads. To enhance accuracy, a consumer-grade composite sensor (including a gyroscope, accelerometer, digital compass, and Kinect v2 depth sensor) was developed. This sensing device is mounted on a robot that automatically rotates around the human subject at an approximate 1-meter radius to capture full-view information. Data streams are processed and fused into a 3D model using an on-board tablet. Additionally, an automatic selection method is proposed to isolate the head portion from the environment. Evaluation against commercial laser scanners (FastSCAN and Cyberware) using Hausdorff distances demonstrates the system's effectiveness and computational efficiency.
Introduction
High-quality 3D head models are essential for fields such as anthropological studies, digital avatar creation, and ergonomic product design (e.g., fitting helmets or glasses). However, the industry faces a "Goldilocks" problem: professional scanners like Cyberware or FastSCAN are highly accurate but prohibitively expensive and require subjects to remain perfectly still. Conversely, consumer sensors like the Microsoft Kinect are accessible but prone to "drift" (accumulative errors) and produce noisy raw data that includes the background and floor.
Before this work, turning a raw scan into a 3D-printable file required significant manual labor to trim away the environment and scale the model correctly. This system seeks to automate the entire "scan-to-print" workflow using a mobile robot and sensor fusion.
System Architecture and Methods
The system follows a multi-stage pipeline from initial sensing to the final physical output.
Hardware Configuration
The sensing device consists of a Microsoft Kinect v2 depth sensor bundled with a custom "composite sensor" (gyroscope, accelerometer, and digital compass). This rig is mounted on a DFRobot HCR Mobile Robot, programmed to circle the subject. A Microsoft Surface tablet mounted on the robot handles the primary computation, while an Arduino Mega manages the robot's movement.
Software and Sensor Fusion
To solve the "drift" problem common in consumer sensors, the system uses a Kalman filter to fuse IMU data with the Kinect’s depth stream. For camera pose estimation, the system assigns a 0.8 weight to the software-based Iterative Closest Point (ICP) algorithm and a 0.2 weight to the hardware sensors. This "hardware-assisted" approach keeps the reconstruction stable even if the software loses track of the subject's features.
Automatic Segmentation and Reconstruction
A key contribution of this work is the automatic selection method. For inanimate models, it uses RANSAC plane fitting to identify and crop the supporting surface. For real humans, it calculates the 3D centroid of the robot’s path to identify the head and creates a virtual bounding prism to extract the bust. The system uses a modified version of KinectFusion to create a Truncated Signed Distance Function (TSDF) volume, which is then converted into a polygon mesh using the Marching Cubes algorithm.
Results and Discussion
The authors evaluated the system using a standard inanimate head model and four real human subjects, comparing results against professional benchmarks. Data from these trials has been published in The MCRLab 3D Human Head Scanning Repository.
- Accuracy: On a standard model, the system achieved a mean Hausdorff distance error of 0.5443 cm. While the professional FastSCAN is more precise (0.1868 cm), the proposed system's error was less than 2% relative to the model's diagonal, which is sufficient for most consumer and ergonomic applications.
- Robustness to Hair: Interestingly, the Kinect’s Time-of-Flight (ToF) approach proved more robust than expensive lasers when capturing dark or shiny hair, which often causes lasers to fail.
- Efficiency: The reconstruction runs at 15 frames per second (near real-time), and physical 3D printing takes between 7 and 12 hours depending on model size.
Conclusion
This research demonstrates a fully automated, cost-effective pipeline for 3D head scanning and printing. By using hardware as a "safety net" for software-based reconstruction, the system achieves professional-grade results with consumer hardware. Future work will focus on adding physical stabilizers (gimbals) to the robot and offloading heavy computation to the cloud to further streamline the process.
Summary updated