Bonemapping: A LiDAR Processing and Visualization Approach and Its Applications

Transcription

1 Bonemapping: A LiDAR Processing and Visualization Approach and Its Applications Thomas J. Pingel Northern Illinois University National Geography Awareness Week Lecture Department of Geology and Geography Eastern Illinois University November 17,

2 Topics What is bonemapping? Simple Morphological Filter (SMRF) Perceptually Shaded Slope Maps (PSSMs) How do we know they work? Empirical testing against hillshaded and hypsometric images, and orthophotos Individual differences What can we use them for? Archaeology at El Pilar Urban applications 2

3 LiDAR involves scanning the environment with a laser sensor that measures distance and direction. When paired with a GPS, we can build a georeferenced point cloud. 3

4 This is a LIDAR-derived Digital Surface Model (DSM) of UC Santa Barbara, represented in shaded relief. 4

5 A first step in LiDAR processing, is the removal of vegetation and/or buildings to get a bare earth representation. 5

6 Terrain Extraction is Important Davidson Library sits approximately 6 meters above the ground due to a terrain layer error. 6

7 Terrain Extraction: The Simple Morphological Filter (SMRF) Emphasizes reducing Earth-as-Object error Still very good at reducing Object-as-Earth error Lowest total error rate of any published algorithm tested against ISPRS dataset tpingel.org/code 7

8 SMRF Workflow 1. Points (x,y,z) are gridded to the minimum value; missing values are inpainted (interpolated) 2. A series of progressive morphological opening operations is conducted with an increasing radius. Each layer is compared with the last; where there is a great deal of change (governed by a slope parameter) that cell is flagged as nonground. 3. At the end, the value of all non-ground cells is inpainted based on the surrounding ground cells. 8

9 A sample progression of SMRF When windowsize = [ ], slope = 15% and elevationthreshold =.5

10 Morphological Operations erosion dilation opening closing 10

11 Cross Section View of Image Opening

12 How well does SMRF perform? SMRF has better performance than other leading algorithms for LiDAR ground filtering Single Parameter (unoptimized) Mean Total Error = 4.4% Axelsson (4.82), Chen (7.23), Shao (4.20) Optimized Mean Total Error = 2.97% SMRF was designed to retain subtle ground cues (minimization of Type I or Earth-as-Object error) for the purpose of visualization. This entails a cost of increased Type II error, which can cause artifacting in the surface.

13 Image Inpainting is applied to fill in the missing cells; assumes that ground returns in the set are common. 13

14

15

16

17 Sometimes, projects have unexpected applications. This is the Ma adim Vallis extraction, based on SMRF. 17

18 The next step is visualization. Here again is a DSM represented as a hillshaded, or shaded relief image. 18

19 This is a Perceptually Shaded Slope Map of the same data. 19

20 Hillshaded images were developed to realistically portray the natural environment at relatively small map scales. Thelin and Pike s (1991) shaded relief map of the United States. 20

21 However, there are significant issues with using shaded relief images to presented LiDAR-derived surfaces. First, more realistic images, like hillshades, do not always perform better than prepared cartographic products. They also do not translate to urban areas very well, where relief is typically low, and aspect is highly regular. 21

22 Kennelly and Stewart (2006) offer some methods that bring out immense detail in a DSM-derived LIDAR visualization. 22

23 But we know that realism is not always best for visualizations. We also need a straightforward way for GIS users to create visualizations with this data that are useful for a variety of purposes, not just more aesthetically pleasing. 23

24 PSSMs are based on the idea of cognitive slope. People exaggerate the vertical component of slope by a factor of 2.3x. (Pingel 2010, following Proffitt et al. 1995) 24

25 Perceptually Shaded Slope Maps (PSSMs) Slope is exaggerated, then mapped to graytone Resulting appearance looks hand-drawn, which speaks to its efficacy as a visualization No spatial displacement errors common with orthophotos Offers a higher contrast image than hillshade, with better affordance for color overlay Most appropriate for mixed / urban environments 25

26 PSSM visualization of a 5 cm LiDAR-derived DSM of Berkeley, CA. 26

27 Empirical Testing of the PSSM Initial tests against 5 visualizations Follow-up test against the shaded relief image Mental Rotation and Profile estimation tasks 27

28 Map Rotation tests saliency. Rotation or rotation + reflection? 28

29 Profile estimation tests legibility. Identify the correct transect from among three alternatives. 29

30 t(176)=1.75, p=0.08 PSSMs showed improved mean accuracy rate and response time on the Profile Estimation task. t(176)=4.58, p<

31 t(175)=3.08, p=0.002 PSSMs showed improved accuracy rates on the Map Rotation task, but no significant improvement on response time. t(175)=1.16, p=

32 % correct Profile Estimation Marginal GLM main effect for sex on response time β=-.89, t(172)=1.70, p=.09 response time (s) 32

33 % correct Map Rotation Interaction between sex and spatial ability on accuracy β=-.08, t(172)=1.99, p=.049 response time (s) 33

34 Last year, we successfully used PSSMs in combination with our LIDAR filter SMRF to visually explore the forest floor of the ancient Maya site, El Pilar. 34

35 Test Case: El Pilar Ancient Maya City on Belize-Guatemala border El Pilar Watering Basin Hundreds of buildings in dozens of plazas Data generously provided by Dr. Anabel Ford 35

36 36

37 37

38 38

39 39

40 Hillshaded image of DSM 40

41 Bonemapping at El Pilar, Guatemala 41

42 42

43 Popular Archaeology article 43

44 44

45 We re currently experimenting with the Velodyne HDL-32E for local area scanning. Compared to airborne LiDAR collection, it boasts higher spatial and temporal resolution. Most applicable for pedestrian use and street capture, although Unmanned Aerial Vehicle mounting is coming soon for our lab.

46

47 The instrument captures scanlines which must be assembled (by the end user!) into a full point cloud.

48 Decimeter resolution rendering of a portion of the Elwood neighborhood in DeKalb, IL. We re currently using bonemapping techniques to visualize the LiDAR data, and to provide accurate tree height estimation.

49 The side view mirrors of vehicles are visible at this resolution.

50 The structure of trees is visible.

51 Our lab is also currently scanning the Northern Illinois University campus. Our goal is to investigate the application of the bonemapping process to interior spaces. 51

52 The Cook County bonemap is available via the CookViewer online as a base layer. Code is available at tpingel.org/code and github.com/thomaspingel. 52

53 Acknowledgements Thanks to many collaborators and student assistants on many of these projects over the past several years. Thanks to the Illinois Geographical Society, the City of DeKalb, and Northern Illinois University for financial support. RealEarth for recent terrestrial LiDAR pointcloud creation 53

54 Selected References Pingel, T.J., Clarke, K.C., and Ford, A. (2015). Bonemapping: A LiDAR Processing and Visualization Technique in Support of Archaeology Under the Canopy. Cartography and Geographic Information Science. 42(S1), [ / ] Pingel, T.J., Clarke, K.C. (2014). Perceptually Shaded Slope Maps for the Visualization of LiDAR Derived Digital Surface Models. Cartographica: The International Journal for Geographic Information and Geovisualization, 49(4), [ /carto ] Pingel, T.J., Clarke K.C., & McBride, W.A. (2013). An Improved Simple Morphological Filter for the Terrain Classification of Airborne LIDAR Data. ISPRS Journal of Photogrammetry and Remote Sensing, 77(1), [ /j.isprsjprs ] 54

55 55