Research Collection
Conference Paper
Quantifying Local Spatial Properties through LiDAR-based Isovists for an Evaluation of Opinion-based VGI in a VR Setup Author(s): Schmid, Konstantin; von Stülpnagel, Rul Publication Date: 2018-01-15 Permanent Link: https://doi.org/10.3929/ethz-b-000225613
Rights / License: Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International
This page was generated automatically upon download from the ETH Zurich Research Collection. For more information please consult the Terms of use.
ETH Library
LBS 2018
http://doi.org/10.3929/ethz-b-000225613
Quantifying Local Spatial Properties through LiDAR-based Isovists for an Evaluation of Opinion-based VGI in a VR Setup Konstantin Schmid & Rul von Stülpnagel Center for Cognitive Science, University of Freiburg, Germany
[email protected],
[email protected]
Abstract. We aim to link VGI that expresses subjective opinions about a location to its spatial configuration. We present an approach to quantify the location’s spatial configuration by generating highly precise isovists based on LiDAR scans. The representativeness of opinion-based VGI and their relation the local spatial configuration can be tested in a VR setup in the lab. Further applications of LiDAR-based isovists are discussed. Keywords. Opinion-based VGI, risk perception, isovists, LiDAR scans
1.
Introduction
The popularity of volunteered geographic information (VGI) has generated a discussion on its quality (e.g., Elwood et al., 2012). This question is further aggravated for opinion-based VGI that cannot be easily matched with an 'objective truth', such as citizens’ reports about perceived risks and hazards during urban cycling (e.g., Nelson et al., 2015): The subjective feeling of risk remains valid even in the absence of an objective hazard. It is not our aim to investigate the preciseness of the VGI’s rather vague geographical indication on a map: The specific location marked as dangerous during urban cycling is less relevant than the described severity as well as the vector of movement resulting in the subjective perception of risk. We rather aim at investigating the representativeness of opinion-based VGI by comparing the content of these contributions with participants’ impressions collected in a controlled lab setting. Furthermore, we aim at linking subjective risk perception to the spatial configuration of a cyclist’s vista space (or viewshed), with the underlying reasoning that spatially complex or rapidly changing situations are experienced as more dangerous. Risk perception
173
LBS 2018
http://doi.org/10.3929/ethz-b-000225613
during urban cycling is affected by a wide range of factors, but an investigation of this aspect has not been achieved so far. For this purpose, we need a method to capture the precise vista space of a cyclist at a given moment. A person’s vista space can be represented in an isovist (Benedikt, 1979), which quantifies the space visible from a given location into a twodimensional polygon, and transfers it into various metric properties (e.g., area, perimeter, drift). Isovist properties of a given point of view have been shown to affect, for example, spatial decision making (Wiener et al., 2007) or the emotional and aesthetic appraisal of buildings (Kuliga et al., 2013). Vice versa, a larger isovist of a landmark’s locations results in an increased selection for sketch-mapping (von Stülpnagel and Frankenstein, 2015). Krukar (2014) demonstrated that a location’s isovist affects visitors’ attention towards and memory for pieces of art. The generation of isovists is, in most cases, based on maps (e.g., Li and Klippel, 2016). However, landscape features clearly affecting visibility (e.g., trees or hedges) are rarely provided in map information (but also see Weitkamp, 2011). In order to investigate the relation between risk perception during urban cycling and the local spatial configuration as captured by isovists, we thus relied on an approach introduced by Dalton and colleagues (2015), who used a LiDAR (Light Detection And Ranging) scanner to generate highly exact isovists of indoor environments. This approach, however, cannot deal with distances exceeding the scanner’s range limits, which is a likely case in outdoor scenarios. Furthermore, transparent or reflective surfaces (e.g., windows) are likely to produce faulty measurements (Yang and Wang, 2008). We are not aware of an available framework suited to provide precise isovists for outdoor spaces. In this paper, we thus introduce an advanced algorithm providing precise isovists of large scale urban landscapes before outlining a study geared towards assessing their relation to risk perception during urban cycling.
2.
Generating Landscape Isovists from LiDAR Scans
A LiDAR scan is recorded by positioning the scanner at the location of interest (i.e., the center of the to-be-created isovist), which provides a point cloud of distance measures in a three-dimensional polar space. Our algorithm reduces the raw 3D point cloud to an isovist in a four-step process, exemplified in Figure 1. As a testbed for the development of our algorithm, we used a Velodyne PUCK - VLP-16 featuring a range from 1m to 100m and a vertical angle of -15° to +15°. Step 1 – Setting vertical channel limits: We generated 2D isovists by using data of only the two most central channels (-1°/+1°) of our scanner. By including more vertical channels, the algorithm could be extended to 3D isovists.
174
LBS 2018
http://doi.org/10.3929/ethz-b-000225613
Step 2 – Setting max values for the isovist radius: The radii of all available data points can be limited depending on the estimated visibility range of the investigated area. This value must not exceed potential limits of the used scanner’s range; in our case 100m. Step 3 – Setting horizontal angle and resolution: All data points of the horizontal angle are binned into a predefined number of equally sized segment bins. The number of segment bins defines the isovist’s resolution. The center of each bin is determined by averaging the azimuth angles and radii of all data points assigned to it. We generated 360° isovists, which are most frequently used. However, the horizontal angle might, for example, be set to 180° to account for a specific direction of view. Step 4 – Removing measuring artifacts: If the scanner cannot detect a light reflection for a given measurement, the respective data point is marked as “not-a-number” (NaN). The number of NaN values was surprisingly high even when using a state-of-the-art LiDAR scanner (e.g., the scan in Figure 1 consists of 24.35% NaN values). Three main reasons can lead to NaN values: (1) The next object exceeds the scanner’s range, (2) the next object consists of a mirroring surface deflecting the laser beam, or (3) the next object is closer than the scanner’s minimal range. Interpolating all NaN values by their next valid neighbors would result in a faulty isovist (see Figure 1, Step 4a). Thus, NaN values were treated on the most probable reason for the faulty measurement. The applied parameters represent preliminary solutions which were based on descriptive inspections of the data and found to yield the most satisfying results. First, we accounted for NaN values resulting from an exceeded scanner range. It can be assumed that relevant open areas (e.g., along a street) continue for several horizontal degrees. Thus, adjacent measurements should yield NaN values as well. Horizontal angle segments of at least 5° with consistent NaN values were treated as open spaces. These were set to the max radius defined in Step 2 (see Figure 1, Step 4b). Segments below the 5° threshold were treated as measuring errors (e.g., due to technical issues), and the respective NaN values were deleted from the dataset. Next, we account for NaN values that were set to the max radius in the previous step, but that resulted either from large mirroring surfaces or objects falling short of the scanner’s range. Larger mirroring surfaces are identified by a software algorithm. Mirroring surfaces (e.g., windows) mostly occur in front of straight-lined objects (e.g., walls). Thus, a regression line for the ten neighboring segment bins below max radius is calculated (see Figure 1, Step 4c). If the mean residual of these neighbors is less than 0.5m, they are assumed to form a straight line. If the examined artifact differs at least 10m from this line, it is assumed to result from a mirroring surface, and the NaN
175
LBS 2018
http://doi.org/10.3929/ethz-b-000225613
Figure 1. Illustration of the four processing steps. Raw laser scan shows an exemplary 3D point cloud. Step 1-3: Initial isovist parameters are applied. Step 4a: Example of a faulty isovist after deleting all NaN values. Step 4b: Data points are set to max when the scanner range is exceeded and binned into segments. (Gray radials indicate segment bins. Only 33% are indicated for reasons of clarity). Step 4c: The red circle marks NaN values caused by the mirroring surface of a car. The neighboring segment bin centers below max radius are part of one straight line (dashed orange line). The NaN values are adjusted to this line (marked with green dots). Generating the final isovist: The laser-based isovist is shown in yellow; the corresponding map-based isovist in grey. Map segment: © OpenStreetMap
value is replaced by a radius adjusted to the straight line. Segments with objects closer than the scanner’s minimal range are identified by another examination of the neighboring segment bins. If one of the neighboring segment bins is closer than 5m to the isovist center, it is assumed that the artifact resulted from a laser ray falling short of the scanner range. The radius of this segment bin is interpolated by a line connecting the centers of the two neighboring segment bins. The interpolated radii are averaged per segment bin, whose centers form the vertices of the isovist. For a preliminary quality check of LiDAR-based isovists computed with our algorithm, we generated an isovist based on OSM information, with buildings as occluding objects and everything else as visible space. The limitations of map-based isovists become obvious on the left bottom part of Figure 1, where obvious visibility obstacles (e.g., trees) remain unaccounted for.
176
LBS 2018
http://doi.org/10.3929/ethz-b-000225613
3.
Application Case: Evaluating Cycling Risks Reported in Opinion-based VGI
We are currently using LiDAR-based isovists to link opinion-based VGI contributions concerning risk perception during urban cycling to the spatial configuration of the respective location. We draw a subset of contributions from a respective VGI project and extrapolated the cycling approach paths resulting in the described hazard. Next, we generated series of LiDARbased isovists and 360° photospheres for these approach paths. Participants wearing an Oculus Rift DK2 studied the successive images and estimated the complexity of each single scene as well as the general hazardousness. In contrast to the presentation of video clips or 2D images, this approach allowed us to probe participants for their impression of a series of particular points of view, while enabling a natural impression of the situation. We are currently investigating how the fluctuation of the vista space (i.e., changing metric properties of the successive isovists, such as their means’ standard deviation or the slope’s trend) affects risk perception when approaching a subjectively dangerous street section, and link them to the semantic severity of the original VGI contribution. One highly interesting measure in this regard would be the isovist’s openness (i.e., the ratio between open and closed edges, Stamps, 2005). Closed edges correspond to objects blocking line of sight , whereas open edges refer to occluded spaces. In the case of urban cycling, open edges reflect the amount of unknown terrain which may hide hazards such as crossing cars. However, laser range scans provide no easy means to allocate edges to one of the two classes without further additional information.
4.
Conclusion
We demonstrated how highly accurate isovists of complex outdoor environments can be generated from LiDAR scans. A systematic comparison of map-based versus LiDAR scan-based isovists was beyond this research, but a preliminary comparison underlined the validity of our approach. The recording of a large number of LiDAR scans remains time-consuming. Our approach is thus most valuable for investigations where precise spatial properties of a limited number of locations are relevant. As was outlined in the previous section, they can be used to provide insights into the representativeness of opinion-based VGI on subjective risks during urban cycling, as well as in the local spatial properties triggering the feeling of subjective risk. This may provide valuable information for urban decision makers aiming to increase the usage of bikes. LiDAR-based isovists as introduced in this paper may also be used to understand, for examples, which
177
LBS 2018
http://doi.org/10.3929/ethz-b-000225613
landmarks support real-world navigation and wayfinding (Schwering et al., 2014), or to investigate the relation of vista space properties and locationbased emotions in pictures uploaded to web platforms (Hauthal and Burghardt, 2013).
References Benedikt, M.L., 1979. To Take Hold of Space: Isovists and Isovist Fields. Environ. Plan. B Plan. Des. 6, 47–65. doi:10.1068/b060047 Dalton, R., Dalton, N., Peverett, I., Clinch, S., Davies, N., 2015. Using laser scanning to produce 3D isovists of real environments. Presented at the 10th International Space Syntax Conference (SSS10). Elwood, S., Goodchild, M.F., Sui, D.Z., 2012. Researching Volunteered Geographic Information: Spatial Data, Geographic Research, and New Social Practice. Ann. Assoc. Am. Geogr. 102, 571–590. doi:10.1080/00045608.2011.595657 Hauthal, E., Burghardt, D., 2013. Extraction of location-based emotions from photo platforms, in: Progress in Location-Based Services. Springer, pp. 3–28. Krukar, J., 2014. Walk, Look, Remember: The Influence of the Gallery’s Spatial Layout on Human Memory for an Art Exhibition. Behav. Sci. 4, 181–201. doi:10.3390/bs4030181 Kuliga, S., Dalton, R.C., Hölscher, C., 2013. Aesthetic and Emotional Appraisal of the Seattle Public Library and its relation to spatial configuration, in: Proceedings of the Ninth International Space Syntax Symposium, Seoul: Sejong University. Li, R., Klippel, A., 2016. Wayfinding Behaviors in Complex Buildings: The Impact of Environmental Legibility and Familiarity. Environ. Behav. 48, 482–510. Nelson, T.A., Denouden, T., Jestico, B., Laberee, K., Winters, M., 2015. BikeMaps.org: A Global Tool for Collision and Near Miss Mapping. Front. Public Health 3. doi:10.3389/fpubh.2015.00053 Schwering, A., Wang, J., Chipofya, M., Jan, S., Li, R., Broelemann, K., 2014. SketchMapia: Qualitative Representations for the Alignment of Sketch and Metric Maps. Spat. Cogn. Comput. 14, 220–254. doi:10.1080/13875868.2014.917378 Stamps, A.E., 2005. Isovists, Enclosure, and Permeability Theory. Environ. Plan. B Plan. Des. 32, 735–762. doi:10.1068/b31138 von Stülpnagel, R., Frankenstein, J., 2015. Configurational salience of landmarks: an analysis of sketch maps using Space Syntax. Cogn. Process. 16, 437–441. doi:10.1007/s10339-015-0726-5 Weitkamp, G., 2011. Mapping landscape openness with isovists. Res. Urban. Ser. 2, 205– 223. Wiener, J.M., Franz, G., Rossmanith, N., Reichelt, A., Mallot, H.A., Bülthoff, H.H., 2007. Isovist analysis captures properties of space relevant for locomotion and experience. Perception 36, 1066–1083. doi:10.1068/p5587 Yang, S.-W., Wang, C.-C., 2008. Dealing with laser scanner failure: Mirrors and windows, in: Robotics and Automation, 2008. ICRA 2008. IEEE International Conference on. IEEE, pp. 3009–3015.
178
