Abstract
Passive stereo imaging is capable of producing dense 3D data, but image matching algorithms generally perform poorly on images with large regions of homogenous texture due to ambiguous match costs. Stereo systems can be augmented with an additional light source that can project some form of unique texture onto surfaces in the scene. Methods include structured light, laser projection through diffractive optical elements, data projectors and laser speckle. Pattern projection using lasers has the advantage of producing images with a high signal to noise ratio. We have investigated the use of a scanning visible-beam LIDAR to simultaneously provide enhanced texture within the scene and to provide additional opportunities for data fusion in unmatched regions. The use of a LIDAR rather than a laser alone allows us to generate highly accurate ground truth data sets by scanning the scene at high resolution. This is necessary for evaluating different pattern projection schemes. Results from LIDAR generated random dots are presented and compared to other texture projection techniques. Finally, we investigate the use of image texture analysis to intelligently project texture where it is required while exploiting the texture available in the ambient light image.
Highlights
Stereo imaging remains a popular technique for dense 3D reconstruction
There have been investigations into algorithmic methods to infer depth in information-poor regions, by far the simplest method to improve match results is to project a random or pseudo-random pattern into the scene. With this approach even simple correlation algorithms are effective; the original Microsoft Kinect device operates on this principle it is not strictly a stereo system (Han et al, 2013). *
Additional stereo pairs were captured with a Kinect pseudo-random dot pattern (KIN) and a random dot image, generated from a data projector (DP)
Summary
Stereo imaging remains a popular technique for dense 3D reconstruction. the performance of stereo matching algorithms is strongly dependent on image texture and scene illumination. There have been investigations into algorithmic methods to infer depth in information-poor regions, by far the simplest method to improve match results is to project a random or pseudo-random pattern into the scene With this approach even simple correlation algorithms are effective; the original Microsoft Kinect device operates on this principle it is not strictly a stereo system (Han et al, 2013). If the LIDAR is cross-calibrated to the stereo system, random dot patterns may be simulated by acquiring dense LIDAR data and projecting the 3D points into each image. This avoids any acquisition bottlenecks caused by the frame rate of the camera.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
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 callMore From: ISPRS - International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
