Abstract
Abstract. Recent trends in 3D scanning are aimed at the fusion of range data and color information from images. The combination of these two outputs allows to extract novel semantic information. The workflow presented in this paper allows to detect objects, such as light switches, that are hard to identify from range data only. In order to detect these elements, we developed a method that utilizes range data and color information from high-resolution panoramic images of indoor scenes, taken at the scanners position. A proxy geometry is derived from the point clouds; orthographic views of the scene are automatically identified from the geometry and an image per view is created via projection. We combine methods of computer vision to train a classifier to detect the objects of interest from these orthographic views. Furthermore, these views can be used for automatic texturing of the proxy geometry.
Highlights
Many 3D scanners provide more information than range data, as for instance color information
Computer vision methods can be applied using these views to train and perform object detection. We present such a work flow that identifies and creates orthographic views. These views are created from registered panoramic images and a proxy geometry that was derived from the point clouds
We presented a work flow for the semiautomatic extraction of orthographic views for indoor scenes from laser range scans and high resolution panoramic images
Summary
Many 3D scanners provide more information than range data, as for instance color information. The scanners are equipped with a camera that takes photographs during the scanning session These photographs are subsequently stitched in order to create a panoramic photo sphere at the scanners position. Such panoramic views can be used to apply color information to pure range data. The problem with wall mounted power sockets is that they usually stick only 3-4 mm out of the surrounding surface This makes them hard to detect in pure range data, as they disappear in the geometric noise of a 3D scan. It is necessary to combine the geometric information of the point cloud scans with the images of the session in order to create orthographic views of planar elements in the scene (e.g. walls) that contain the desired perceived objects (e.g. sockets). These views are created from registered panoramic images and a proxy geometry that was derived from the point clouds
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
