In the field of optical 3D scanning for healthcare applications, low-cost depth cameras can be efficiently used to capture geometry at video frame rates. However, the complete reconstruction of anatomical geometries remains challenging since different scans, collected from multiple viewpoints, must be aligned into a common reference frame. This paper proposes a fully automatic procedure to align scans of the upper limb patient’s anatomy. A 3D optical scanner, obtained by assembling three depth cameras, is used to collect upper limb acquisitions. A relevant dataset of key points on the hand and the forearm geometry is then determined and used to automatically obtain a rough 3D alignment of the different scans. Hand key points are identified through a neural network, which works on RGB images captured by the depth cameras; forearm key points are recognized by directly processing the point clouds through a specifically designed algorithm that evaluates the skeleton line of the forearm. The approach was tested on forearm acquisitions, and the results were compared to alternative alignment methodologies.
File in questo prodotto:
Non ci sono file associati a questo prodotto.