Accelerating Point Cloud Cleaning
| dc.contributor.author | Mulder, Rickert L. | en_US |
| dc.contributor.author | Marais, Patrick | en_US |
| dc.contributor.editor | Chiara Eva Catalano and Livio De Luca | en_US |
| dc.date.accessioned | 2016-10-05T06:27:48Z | |
| dc.date.available | 2016-10-05T06:27:48Z | |
| dc.date.issued | 2016 | |
| dc.description.abstract | A laser scanning campaign to capture the geometry of a large heritage site can produce thousands of high resolution range scans. These must be cleaned to remove noise and artefacts. To accelerate the cleaning task, we can i) reduce the time required for batch-processing tasks, ii) reduce user interaction time, or iii) replace interactive tasks with more efficient automated algorithms. We present a point cloud cleaning framework that attempts to improve each of these aspects. First, we present a novel system architecture targeted point cloud segmentation. This architecture represents 'layers' of related points in a way that greatly reduces memory consumption and provides efficient set operations between layers. These set operations (union, difference, intersection) allow the creation of new layers which aid in the segmentation task. Next, we introduce roll-corrected 3D camera navigation that allows a user to look around freely while reducing disorientation. A user study showed that this camera mode significantly reduces a userĀ“s navigation time between locations in a large point cloud thus accelerating point selection operations. Finally, we show how boosted random forests can be trained interactively, per scan, to assist users in a point cleaning task. To achieve interactivity, we sub-sample the training data on the fly and use efficient features adapted to the properties of range scans. Training and classification required 8-9s for point clouds up to 11 million points. Tests showed that a simple user-selected seed allowed walls to be recovered from tree and bush overgrowth with up to 92% accuracy (f-score). A preliminary user study showed that overall task time performance was improved. The study could however not confirm this result as statistically significant with 19 users. These results are, however, promising and suggest that even larger performance improvements are likely with more sophisticated features or the use of colour range images, which are now commonplace. | en_US |
| dc.description.sectionheaders | Acquisition and Processing | |
| dc.description.seriesinformation | Eurographics Workshop on Graphics and Cultural Heritage | |
| dc.identifier.doi | 10.2312/gch.20161410 | |
| dc.identifier.isbn | 978-3-03868-011-6 | |
| dc.identifier.issn | 2312-6124 | |
| dc.identifier.pages | 211-214 | |
| dc.identifier.uri | https://doi.org/10.2312/gch.20161410 | |
| dc.identifier.uri | https://diglib.eg.org:443/handle/10.2312/gch20161410 | |
| dc.publisher | The Eurographics Association | en_US |
| dc.subject | I.3.1 [Picture/Image Generation] | |
| dc.subject | Digitization and Image Capture | |
| dc.subject | Scanning | |
| dc.title | Accelerating Point Cloud Cleaning | en_US |
Files
Original bundle
1 - 1 of 1