Abstract
Abstract. The recording of high resolution point clouds with sub-mm resolution is a demanding and cost intensive task, especially with current equipment like handheld laser scanners. We present an image based approached, where techniques of image matching and dense surface reconstruction are combined with a compact and affordable rig of off-the-shelf industry cameras. Such cameras provide high spatial resolution with low radiometric noise, which enables a one-shot solution and thus an efficient data acquisition while satisfying high accuracy requirements. However, the largest drawback of image based solutions is often the acquisition of surfaces with low texture where the image matching process might fail. Thus, an additional structured light projector is employed, represented here by the pseudo-random pattern projector of the Microsoft Kinect. Its strong infrared-laser projects speckles of different sizes. By using dense image matching techniques on the acquired images, a 3D point can be derived for almost each pixel. The use of multiple cameras enables the acquisition of a high resolution point cloud with high accuracy for each shot. For the proposed system up to 3.5 Mio. 3D points with sub-mm accuracy can be derived per shot. The registration of multiple shots is performed by Structure and Motion reconstruction techniques, where feature points are used to derive the camera positions and rotations automatically without initial information.
Highlights
The recording of high resolution geometry data in cultural heritage applications is a challenging task, especially for large scale objects
Handheld laser scanners often suffer of low acquisition efficiency and of orientation problems, while the costs for such systems are very high
Natural materials like stone or wood provide good texture so that each pixel has a distinct grey value. This is important since the image matching accuracy is reduced if two neighboring pixels have the same value
Summary
The recording of high resolution geometry data in cultural heritage applications is a challenging task, especially for large scale objects. At once, image based systems can provide 3D surface information for a whole area due to the matrix structure of the image This is beneficial for difficult environments where only a small acquisition distance is possible or where vibrations or movements occur. If a point is observed in two 2D images from different views, its 3D coordinate can be reconstructed by intersecting the viewing rays This method can be extended to multiple images, where the ray intersection provides higher accuracy and high reliability. Feature based image matching methods are extracting distinct features like points or lines, which are identifiable in multiple images This approach is especially suitable for the reconstruction of 3D information without initial information like applied in Structure and Motion reconstruction methods, where camera pose and sparse surface information are automatically derived from images. We use only this projected pattern to support our image matching, while omitting the lowresolution depth information computed by the Kinect
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
