TRAIL2014 - Workshop LiDAR data processing

July 4, 2017 | Autor: Catherine Fruchart | Categoria: Archaeology, Remote Sensing, 3D visualisation, LiDAR for Landscape Archaeology
Share Embed


Descrição do Produto

The remains of a medieval castle are situated in the red box below. We need to produce 2D and 3D images of this area, to understand the global organization of the historic site and to evaluate its total extent and also to produce some images to illustrate the archaeological report. The existing DTMs have provided bad results because of point cloud filtering issues. To solve this problem, we can realize a new DTM by using a method that doesn’t use algorithmic points filtering. Basically, points filtering is difficult here because of: - topography: cliffs, rugged terrain - woodland cover: thick persistent low and medium height vegetation -anthropogenic context: urbanized environment The ruined castle was settled here at the edge of a cliff

C. Fruchart 2014 – LIEPPEC / MSHE C.N. Ledoux

Snapshot of the place: view of the road at the foot of the cliff on Google Maps.

medieval ruined castle here

3D snapshots of the place and its surroundings on Google Earth: cliffs are badly rendered .

N

N castle here

This is a 2D view of the area derived from the first DTM that was realized (visualization = simple hillshade). This 50 cm resolution DTM was provided by the company in charge of the LiDAR survey in 2009 (discrete multi-echo return, average ground points density = 4-5 pts/m2). Points filtering was performed with Terrascan using a customized algorithm designed by engineers in charge of the job and the DTM was realized with Global Mapper from a TIN model based on a Delaunay N triangulation. This DTM has given good results for the major part of the surveyed area, but is not good for this area.

The castle is here

C. Fruchart 2014 – LIEPPEC / MSHE C.N. Ledoux

about filtering: http://www.opentopography.org/index.php/blog/detail/tools_for_lidar_point_cloud_filtering_classification Axelsson, 2000: http://www.isprs.org/proceedings/XXXIII/congress/part4/111_XXXIII-part4.pdf Sithole, Vosselman, 2003 http://www.itc.nl/isprswgiii-3/filtertest/MainDoc.htm about Delaunay triangulation: http://www.princeton.edu/~achaney/tmve/wiki100k/docs/Delaunay_triangulation.html http://web.pdx.edu/~jduh/courses/geog493f09/Students/W6_DelaunayTriang.pdf )

The 3D visualization of the area realized with this first DTM is not exploitable.

C. Fruchart 2014 – LIEPPEC / MSHE C.N. Ledoux

This is a view of the LiDAR pointcloud around the castle

castle here

A

B

C. Fruchart 2014 – LIEPPEC / MSHE C.N. Ledoux

A

B

A manual point cloud filtering of the zone is difficult to realize, partly because the surface to filter is fairly large - at least 400 m x 50 m – and mostly because the ground is a rugged terrain covered with thick and persistent low vegetation. Therefore, it is very hard to decide which points belong to the low vegetation and which ones belong to the ground.

A new model with a 60 cm resolution has been realized to get a better image of the zone (shaded view below). As a principle, points are not classified into categories, and each cell takes the Z value that corresponds to the minimum Z value among all the points within each cell area extent (in this example, 60 cm x 60 cm). This very simple way of generating a digital model is usable here because the density of the LiDAR pointcloud is big enough to get, for most of the cells, one point actually belonging to the ground and because the observation scale we are using and the features we are studying for an archaeological purpose are observed at a size corresponding to human scale.

N

castle here

C. Fruchart 2014 – LIEPPEC / MSHE C.N. Ledoux

In this example, because low vegetation was very thick, 2 intermediate models were realized at first, each with a resolution of twice the final expected resolution (= 1,2 m here for a final 60 cm resolution). One of the two intermediate model was processed with a grid with X,Y coordinates shifted from 60 cm from the first grid. Thanks to that, we can get remove more low vegetation points and eventually, after having interpolated these 2 intermediate 1,2 m resolution models, get a final model with a 60 cm resolution and few noise from low vegetation.

Same process with a 40 cm final resolution (shaded model below): low vegetation noise is more visible but sharpness is better. For visual purpose or aesthetic issues (publication etc.), different models can be superimposed on a bunch of layers and mixed up with appropriate software to keep the best of each layer (e.g. Photoshop).

N

C. Fruchart 2014 – LIEPPEC / MSHE C.N. Ledoux

Shaded view with first available DTM

C. Fruchart 2014 – LIEPPEC / MSHE C.N. Ledoux

Shaded view with new model

A 1m resolution model was realized for the archaeological report to get rid as much as possible of low vegetation. Buildings are ugly and relief details lack in accuracy, but archaeological features appear clearly and with minimum disturbing noise.

3D view with 40cm resolution shaded model

C. Fruchart 2014 – LIEPPEC / MSHE C.N. Ledoux

This method can also be used to perform a DEM that allows a better3D photographic visualization than the one available on Google Earth.

C. Fruchart 2014 – LIEPPEC / MSHE C.N. Ledoux

N

N

Lihat lebih banyak...

Comentários

Copyright © 2017 DADOSPDF Inc.