Automatic Multiview Alignment of RGB-D Range Maps of Upper Limb Anatomy

Area of Science:

• Biomedical Engineering
• Computer Vision
• Medical Imaging

Background:

• Accurate 3D anatomical models are vital for biomedical applications.
• Existing multi-camera systems for 3D scanning are often expensive and bulky.
• Forearm 3D scanning presents challenges like multiple views, stability, and optical undercuts.

Purpose of the Study:

• To develop an automatic alignment procedure for creating accurate 3D models of upper limb anatomy.
• To assess the feasibility of using multiple consumer RGB-D sensors for anatomical scanning.
• To overcome challenges in forearm 3D scanning using a low-cost handheld device.

Main Methods:

• A handheld scanner with three Intel RealSense D415 depth cameras was assembled.
• Automatic alignment involved extracting common key points using a neural network and a custom skeleton line algorithm.
• Fine registration utilized a specifically developed iterative-closest-point (ICP) algorithm.

Main Results:

• The proposed method achieved deviations below 5 mm, with a mean of 1.5 mm, in forearm scans.
• The automatic alignment procedure successfully registered point clouds from different scanner poses.
• The system demonstrated significant improvements compared to manual alignment and existing methods.

Conclusions:

• The developed approach accelerates 3D acquisition and automatic registration of upper limb anatomy.
• This method offers a cost-effective alternative to expensive multi-camera systems.
• The findings support the development of personalized biomedical applications and upper limb rehabilitation frameworks.