Visualization of Activity Data from a Sensor-based Long-term

Monitoring Study at a Playground

Tobias Bolten

, Regina Pohle-Fr

ohlich

, Dorothee Volker

, Clemens Br

uck

, Nicolas Beucker

and Hans-G

unter Hirsch

Institute of Pattern Recognition, Niederrhein University of Applied Sciences, Reinarzstr. 49, Krefeld, Germany

Competence Center for Social Urban Design, Niederrhein University of Applied Sciences,

Frankenring 20, Krefeld, Germany

Keywords:

Activity Visualization, Heat Map, Long-Term Monitoring, Dynamic Vision Sensor.

Abstract:

In the context of urban planning, a detailed knowledge of the considered space and its utilization is essen-

tial. However, manual observations are often not performed due to cost. Whereas sensor-based systems are

often not installed due to possible constrains caused by data protection laws and user privacy-concerns. We

addressed these concerns and developed a privacy-aware, sensor-based processing pipeline for detecting ob-

jects based on an analysis of signals from several sensors. These detections are used for their mapping and

visualization in a global bird eye view.

Besides a data normalization, which is crucial considering sections of different lengths, multiple variations of

activity visualization applying heat maps are described. This includes the utilization of background represen-

tations with different levels of details, different accumulations of object detections through the adjustment of

the performed spatial binning as well as applying different colormaps. Both sequential colormaps and diverg-

ing colormaps with and without perceptually uniform distances were considered.

These variations were evaluated in a conducted online survey addressing professional urban planners as well

as interested citizens. The results of this survey were used to determine a meaningful default setup for visual-

izing the activities in an interactive graphical user interface. This interface is intended to make the results of

the performed long-term monitoring generally accessible.

1 INTRODUCTION

In order to realize an user-oriented perspective for

planning and designing public spaces, urban planners

need detailed knowledge about the future or current

users of an urban space. For this reason, a long-

term observation of the behavior of the different user

groups in the public space is necessary.

Until now the analysis of interaction in public

space is generally still based on analogue methods,

which leads to rather expensive user studies. Man-

ual methods are mainly preferred because of ethical

standards concerning pattern recognition and the sur-

plus of ad hoc interpretation by professional observers

(Gehl and Svarre, 2013). Depending on very differ-

ent international legislations, monitoring by means of

classic CCTV systems is often associated with high

regulatory requirements. Regardless of the legal con-

straints, these CCTV systems can also be associated

with user concerns regarding data protection aspects.

In this paper, we present generated activity vi-

sualizations resulting from a sensor-based long-term

study conducted in a living lab context of a pub-

lic children’s playground. Within this study, the

aforementioned user concerns were considered and

addressed through an alternative, privacy-compliant

sensor selection. The goal is to use these evidence-

based activity observations as a resource to evaluate

spatial use and optimize urban design considerations.

Due to the long-term aspect of this monitoring and

the utilization of multiple sensors, a large amount of

data is generated in this process. Therefore, it is nec-

essary to visualize it in a suitable manner. The key is a

presentation in a way that decision makers can under-

stand and interpret the diverse collected information.

For this purpose, we are examining several aspects of

activity visualization and investigate them using ex-

amples through a conducted user study among profes-

sional decision makers in the urban planning environ-

ment. Furthermore we are exploring the visualization

146

Bolten, T., Pohle-Fröhlich, R., Volker, D., Brück, C., Beucker, N. and Hirsch, H.

Visualization of Activity Data from a Sensor-based Long-term Monitoring Study at a Playground.

DOI: 10.5220/0010798700003124

In Proceedings of the 17th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2022) - Volume 3: IVAPP, pages

146-155

ISBN: 978-989-758-555-5; ISSN: 2184-4321

of various data based insights in one interface in or-

der to generate a user friendly overview of complex

situations.

The paper is structured as follows. Section 2 sum-

marizes related work in the context of activity visual-

ization. A detailed description of the monitored area

and the processing pipeline used for activity detection

follows in Section 3. In Section 4 the investigated

visualizations, as well as their creation is described.

First results of the user study are summarized in Sec-

tion 5. Finally, an overall summary and outlook for an

integrated user friendly interface is provided in Sec-

tion 6.

2 RELATED WORK

There are different approaches for the visualization of

space-time data in the literature. For example, mea-

sured parameters can be used to color individual ob-

jects of the observed infrastructure. However, the ob-

served playground in this study is designed as a nature

experience space and therefore does not contain clas-

sic playground equipment, but rather different areas

in which the children can playfully interact with nat-

ural materials. Therefore, in this practical case, play-

ground equipment representations cannot be used for

visualization.

More popular is the visualization of the measured

values from the user’s perspective by creating usage

maps. It can be observed that abstract 2D represen-

tations are preferred compared to 3D representations.

This is certainly also due to the fact that 2D repre-

sentations do not require any additional interaction

when interpreting the data. In some projects, such as

(Rezaei and Azarmi, 2020), both options are used to

visualize tracking and distance maps.

In addition to conveying the acquired data by

means of coloring as 2D or 3D heat map, there

are also many examples on the web that use simple

shapes such as circles, spheres, or cubes to represent

site-speciﬁc information (Geospatial, 2021). Depend-

ing on the type of available data and the desired type

of information to be conveyed, human movements in

public spaces are visualized in different ways. For

example, to visualize the paths of individual visitors,

Cuellar chooses a single ﬂow representation (Cuel-

lar et al., 2020), while Laureyssens chooses non-

personalized ﬂow maps (Lauressens, 2005) to display

pedestrian lanes in a public square. In the ﬂow map in

(Nielsen et al., 2014), it is possible to assign the ﬂow

lines to individual moving people. An extension of

ﬂow maps can be found in (Peysakhovich and Hurter,

2017). Here, in addition to the path of the movement,

the direction of the movement is also used in a ﬂow

direction map to display the results of eye tracking.

Scatterplots are used to visualize fewer recorded

events at a location, such as showing manually

counted people in a street segment within a limited

time period. This is for example described by Whyte

in (Whyte, 1979), who was a pioneer of systematic

user analysis in urban planning. Very often, heat maps

can also be found to represent captured people when

automatic data collection is performed, e.g., in (Ra-

jasekaran et al., 2020) to represent group activities

in student dormitories. In (Rashed et al., 2016) heat

maps are used to analyze which exhibit in a museum

is particularly popular for visitors. Bolleter also uses

heat maps to present his results of Wi-Fi tracking in

public spaces to count people and map their stays and

movements at the district level (Bolleter, 2017).

When using heat maps, it should be noted that the

selection of an appropriate color table and a scaling

adapted to the problem are crucial for the subsequent

interpretation of the data (Eghteabs et al., 2017). In

addition, when selecting the color tables, it is impor-

tant to consider that color-blind people can also inter-

pret the data and that older people are less sensitive to

colors (Silva et al., 2011).

Different types of generation for colormaps are

described in the literature (Zhou and Hansen, 2015).

Colormaps can be generated procedurally, with the

goal of being able to interact with as many different

data sets as possible. For some tasks, user studies ex-

ist that were used to develop suitable colormaps. Fur-

thermore, perception-based rules learned in the course

of life exist for certain applications. An example of

this is the communication of the measured tempera-

ture via a blue-red scale. For our monitoring applica-

tion, the data-driven generation is relevant to make

both, the short-time and the long-term data, inter-

pretable in a similar way with an adapted colormap.

For visualization of such ordinal data, sequential per-

ceptually uniform maps should be used that reﬂect nu-

merically equal distances between values for equally

perceived color differences (Moreland, 2009).

3 DATA ACQUISITION AND

PROCESSING

This section presents the developed processing chain

from the activity recording performed by multiple

sensors to the derived coordinates for a joint visual-

ization.

Visualization of Activity Data from a Sensor-based Long-term Monitoring Study at a Playground

147

Figure 1: Schematic plan of the measuring area and sensor

positioning (adapted from (Bolten et al., 2021b)).

3.1 Living Lab

The schematic plan of the children’s playground,

which serves as a study area within the living lab, is

shown in Figure 1. This area has an approximate size

of 2800 square meters and is covered by three perma-

nently mounted sensors. The recording scenario of

this long-time observation corresponds to the techni-

cal setup of (Bolten et al., 2021b) and can be summa-

rized and quoted as follows.

Each of the sensors is mounted on a pole at a

height of approximately 6 meters with an angle of in-

clination of approximately 25 degrees to the ground.

The positioning of the sensors was chosen in such

a way that blind spots created by the terrain (trees,

bushes, hills) are minimized as far as possible. Fur-

thermore, it was tried to cover all areas by more than

one sensor. The recording period of each sensor cov-

ered the timespan from 10am to 8pm a day all week,

which results in a very large amount of data that

serves as a basis for visualization.

3.2 Privacy-aware Activity Data

Collection

Activity on the playground was captured using three

CeleX4 Dynamic Vision Sensors (DVS) (Guo et al.,

2017), each equipped with a 8mm wide-angle lens.

Compared to conventional cameras, these sensors do

not record intensity frames, but capture only changes

in brightness as a sparse stream of so called events.

Each pixel of a DVS operates independently and

asynchronously from each other and triggers an out-

put event when a brightness change above a thresh-

old has been detected. An event e = {(x , y),t, p} con-

tains only the spatial position (x,y) of the triggered

pixel, a timestamp t and the polarity p of the bright-

ness change.

Figure 2: Semantic segmentation result of DVS event

stream with derived object detection coordinate S

(x,y).

These types of vision sensors offer a fundamental

advantage in terms of data privacy aspects, as no gray

or color values need to be processed in any software

logic to evaluate the activity within the scene. These

inherent hardware-based sensor characteristics allow

DVS technology to be used under signiﬁcantly lower

legal data protection requirements compared to tradi-

tional CCTV systems. Furthermore, the DVS tech-

nology also offers reduced data redundancy through

the data-driven output paradigm of the sensor, as well

as a higher dynamic range, lower power consumption

and higher temporal resolution compared to classic

CCD/CMOS imagers. The utilized CeleX4 sensor of-

fers a spatial resolution of 768 × 640 pixels, of which

768 × 512 pixels are effectively used due to technical

limitations.

To analyze the event stream, an event-wise seman-

tic segmentation is carried out for contiguous tempo-

ral windows of the data streams, each of 60ms du-

ration. This segmentation is achieved by applying a

PointNet++(Qi et al., 2017) deep neuronal network

structure, which is especially designed for the pro-

cessing of unordered point clouds. Thus, each event

is assigned a label out of a set of 10 classes in to-

tal. In addition to the classes person, dog, bicycle

and sportsball, which represented objects of interest

in the performed monitoring, six classes for environ-

mental inﬂuences, such as bird, insect, tree crown,

tree shadow, rain, and shadows were considered. This

segmentation step is based on (Bolten et al., 2021a)

and their published models. According to the evalua-

tion provided in this paper, the PointNet++ processing

achieves an occurrence frequency weighted F1-score

of ≈ 93% over all mentioned classes on the DVS-

OUTLAB dataset (Bolten et al., 2021b), which was

recorded in the same application context as this work.

Within the segmented event stream, the events of

classes of interest are clustered and thereby the detec-

tion coordinates S

(x,y) relative to ﬁeld of view from

sensor S are determined. As an example, a detected

person is shown in Figure 2.

In addition to the data collected with the DVS,

IVAPP 2022 - 13th International Conference on Information Visualization Theory and Applications

148

(a) 3D Rendering (b) Wireframe

Figure 3: Calculated SFM 3D model of the playground.

Figure 4: 3D model based coordinate lookup table genera-

tion.

weather data was also recorded to be included in the

planned interface as illustrated in Figures 13. This

data can contribute to a better estimation of the use of

the play area at a certain time, as fewer users can be

expected during cool rainy weather than during warm

sunny weather.

3.3 3D Model-based Bird’s Eye View

Mapping

The next step within the processing pipeline used for

visualization is to map the object detections S

(x,y),

which are relative to the sensor ﬁeld of view, into a

global representation such as a top or blueprint view

like Figure 1.

The mapping of these object detection coordinates

is not a trivial task due to the heavy terrain modeling

(like hills and valleys) of the monitored playground

area. Due to these terrain properties, a simple afﬁne

transformation between the sensor ﬁeld of view and

the map representation leads to very large inaccura-

cies, since no ﬂat surface plane is given.

To achieve a more accurate estimate, we cal-

culated a 3D model of the playground area, and

performing the mapping based on this model. In this

way, the terrain modeling is included in the applied

transformation. This 3D model was generated using

285 images from a drone overﬂight. In these overlap-

ping images, the camera positions are estimated based

on determined corresponding points using structure

from motion (SFM) technique. Based on these

camera positions, the 3D points of the model are cal-

culated. We used the commercial software AgiSoft

for this purpose. A 3D rendering of the created model

https://www.agisoft.com/

Figure 5: Uncertainty visualization of performed S

(x,y) to

) point mapping, exemplary for one sensor.

Figure 6: Cropped and zoomed example of uncertainty

based merging of projected bird eye detections.

is shown in Figure 3a, where Figure 3 highlights the

included terrain characteristics of the model.

Afterwards the calculated 3D model of the play-

ground is used in a 3D simulation created with

Blender

. Within this simulation, camera objects

were positioned and aligned with the same parameters

as the used DVS sensors (such as sensor size, resolu-

tion, focal length). In addition, a bird eye camera B

was positioned. Through this simulation, the relation

of the 3D model mesh vertices to the respective pixel

coordinates in the DVS view or bird eye view can be

determined. In this way, a lookup table was gener-

ated that allows a mapping from S

(x,y) to B

)

within the rendered top view. This processing is illus-

trated in Figure 4.

3.4 Fusion of Detections from Multiple

Sensors

The described method achieves a high quality map-

ping between the (x,y) coordinates in the sensor view

to corresponding points in the bird eye top view.

However, due to the terrain modeling, there are still

leaps in this bird eye view mapping. Figure 5 shows

https://www.blender.org/

Visualization of Activity Data from a Sensor-based Long-term Monitoring Study at a Playground

149

this as an example for one of the sensors. Here the

maximum distance in the bird eye projection is shown

for each point S

(x,y) of the sensor view, which re-

sults when projecting the coordinates within the 8-

connected neighborhood S

(x ± 1,y ± 1) around this

point. On the ridges of the terrain modeling projec-

tion distances of several meters can result by shifting

the sensor detection by only one pixel. This distance

is taken into account as an “uncertainty factor” when

merging the detections from all sensors into a joint

representation.

For temporal synchronous detections from differ-

ent sensors, it is tested whether there are overlaps to

other detections within these uncertainty distances.

This procedure is illustrated in Figure 6. In this ex-

ample, the ﬁlled circles correspond to synchronous

detections from the sensors and the outer circles in-

dicate the corresponding uncertainty distances of the

bird mapping. The radii of the detections of DVS1

and DVS3 overlap in this case and are therefore com-

bined into one detection.

4 STUDIED VISUALIZATIONS

In the following, the accumulation logic for a con-

solidated visualization of the merged object detec-

tions is presented. This includes data binning as

well as normalization. Subsequently, different gener-

ated and studied visualizations variants are presented.

All generated activity plots were created using the

matplotlib library(Hunter, 2007).

4.1 Spatial- and Temporal Detection

Binning

In order to visualize the activity level on the play-

ground area, the bird eye view is divided into several

sub-areas in the form of a spatial binning. For each

subarea, the activity is then determined and plotted.

The grid for this spatial partitioning is approximately

0.5 to 1 square meter per bin. This selection was made

in order to be able to identify slight location differ-

ences in the usage-frequency despite the overall size

of the monitored area.

Within the performed long-term observations,

there are periods of very different lengths in which an

activity has taken place on the site. Since both short-

time and long-term aspects are of interest within the

evaluation, the ability to visualize time intervals of

different lengths is essential. This results in the chal-

lenge that the displayed activity levels (low to high)

must be comparable with each other when visualizing

periods of different lengths by using the same color

values. For this reason, we apply the following nor-

malization technique.

In addition to the described spatial binning of

the observed area, a further temporal binning is per-

formed. For each spatial bin, the detections of that

area are sorted temporally. The complete temporal in-

terval considered for visualization is then divided into

smaller sections and the count of these smaller tempo-

ral sections containing at least one object detection is

determined. In this way, a percentage normalization

between [0,1] is realized as a function of the consid-

ered total duration.

For example, a considered observation time for vi-

sualization of one hour and a division into temporal

bins of one minute length results in a scaling of x/60,

with x being the number of populated one minute de-

tection time blocks. The division into these short time

segments (e.g. one minute) is possible because the se-

mantic analysis of the DVS event stream is performed

in much shorter time windows of only 60ms duration,

as explained in Section 3.2.

4.2 Considered Heat-map Variations

Then the merged, binned and normalized object de-

tection data is visualized in the form of heat maps to

give decision-makers an intuitive and quick overview

and to make signiﬁcant areas of usage easily recog-

nizable.

Different variations in the generation of these heat

maps were considered and evaluated:

Level of Detail in Bird Eye View (LoD)

For a spatial orientation of the visualized activity

detections, a representation of the monitored area

is crucial. However, the background image used

as site representation can be carried out in differ-

ent levels of detail.

Photorealistic rendered variants from the created

3D Blender model simulation, as well as further

abstracted sketches of the considered area were

evaluated in our work. Examples of these back-

ground image variants in color, as well as in

grayscale, are shown in Figure 7.

2D vs 3D Display Dimensionality

Display variants with a 2-dimensional and a 3-

dimensional rendering of the plot were created.

In the case of the 3D display, the activity is rep-

resented by the height of a three-dimensional sur-

face in addition to the used color. An example of

this is given in Figure 8a. Since a change of the

viewport is necessary to interpret these 3D views,

animations were created containing a 360 degree

rotation of the plot.

IVAPP 2022 - 13th International Conference on Information Visualization Theory and Applications

150

(a) RGB Rendering (b) Greyscale

rendering

sketch

(e) Simpliﬁed

sketch

Figure 7: Used level of detail in bird eye view.

(a) 3D with animated viewport (b) 2D projection

Figure 8: Spatial display variants in 2D and 3D.

(a) Rectangles (b) Hexagons (c) Contour (d) Continuous density, result-

ing from KDE

Figure 9: Applied spatial binning and display variants (highlighted region from Figure 8b).

(a) Bone (b) BuGn (c) RdPu (d) YlOrBr

Figure 10: Applied sequential colormaps.

A representation as a two-dimensional image, on

the other hand, allows the data to be interpreted

without this type of animation or interaction. A

corresponding example for this visualization is

given in Figure 8b.

Spatial Binning Variants

Besides the straightforward variant of the spatial

(x,y)-binning into rectangular tiles we also ap-

plied a tessellation by regular hexagons, as dis-

played in Figures 9a and 9b. In addition, a

contour-based representation was implemented,

which shows areas with approximately similar us-

age appropriately grouped together (Figure 9c).

The variants mentioned so far have a discontinuity

of the generated representation in common. To

provide an alternative for this, a Kernel-Density-

Visualization of Activity Data from a Sensor-based Long-term Monitoring Study at a Playground

151

(a) Viridis (b) Jet (c) Gnuplot2 (d) Bwr

Figure 11: Applied diverging colormaps.

Estimation (KDE) based variant was also created,

which allows a gradually smooth representation

of the activity level. An example of this is given

in Figure 9d.

Colormap Variations

Compared to long-learned and daily applied col-

ormaps (e.g. in weather forecasts), the question is

which colormap is best suitable, interpretable and

to intuitively associated with a level of activity by

the addressed decision makers.

For this purpose, various simple sequential col-

ormaps (Figure 10), as well as further diverging

colormaps (Figure 11) were examined. In case

of the diverging colormaps, contrary to usual use-

cases, the common application of a center-point

within the double-ended color gradient was omit-

ted. This results in a continuous color fading of

the normalized activity level from low to high

through this center point.

Perceptually uniform perceived colormaps were

also taken into account in this colormap consid-

eration. In addition the direction of the used col-

ormap to encode the activity level, i.e. from dark

to bright or vice versa, were also considered.

A linear alpha-blending with higher transparency

factors for lower activity levels was used within

the colormap to overlay the generated activity vi-

sualizations over the selected plot background im-

age.

5 USER STUDY

The previously described representations of the ac-

tivity heat maps were evaluated in an online survey.

The carefully selected participants came from profes-

sional backgrounds in urban planning and interaction

design, but also interested citizens were involved in

order to reﬂect the layman’s perspective. The sur-

vey was designed to check the comprehensibility of

(a) 3 (b) 5 (c) 7 (d) 255

Figure 12: Used level of detail in activity colormap through

further binning it into n blocks.

the generated activity representations by heat maps,

which were created based on design assumptions re-

garding readability and experiential knowledge, and

to identify adequate default settings for upcoming vi-

sualizations and applications. Sixteen participants

took part in the online survey. They had the opportu-

nity to rate the different visualizations and were asked

to comment their decisions.

The results show trends and conﬁrm previously

made assumptions. Table 1 summarizes numerically

the survey results described below.

5.1 Visualization of the Playground

The RGB rendering of the bird eye image (Figure 7a)

was selected as the preferred variant. But the com-

ments revealed that the displayed user activity could

not be related precisely as the heat maps covered the

image with solid colors.

Regarding the 3D rendering, it can be stated that

a high user activity is displayed with a large distance

from the map, which makes the assignment more dif-

ﬁcult.

5.2 Colormaps and Binning

In terms of color, it can be summarized that color

maps with diverging colors are preferentially associ-

IVAPP 2022 - 13th International Conference on Information Visualization Theory and Applications

152

Table 1: Summarized overview of response rates of the user survey.

Aspect of question

Options

RGB

Rendering

Grey

Rendering

RGB

Sketch

Grey

Sketch

Simpl.

Sketch

None

LoD Bird eye view (Fig. 7) 9 0 1 2 3 1

3D ani-

mated

2D RGB 2D Grey

2D RGB

Sketch

2D Grey

Sketch

2D Simpl.

Sketch

3D / 2D representation (Fig. 8) 0 9 1 1 4 1

Rectangles Hexagons Contour KDE

Spatial binning (Fig. 9) 7 1 3 5

Bone BuGn RdPu YlOrBr None

Sequential colormaps (Fig. 10) 3 5 4 2 2

Viridis Jet Gnuplot2 Bwr

Diverging colormaps (Fig. 11) 2 9 3 2

3 Blocks 5 Blocks 7 Blocks Full

Colormap binning (Fig. 12) 2 0 4 10

ated with user activity. One assumption why these

diverging colormaps were preferred over the sequen-

tial ones is that they offer a higher contrast and are

thus, also for older users, more clearly distinguish-

able. This contrast discrepancy is further enhanced

by the ampliﬁed alpha blending factor used at lower

activity levels (compare to Section 4.2).

However, contrary to the widespread assumption

from the literature (Moreland, 2009), no perceptually

uniform colormap was preferred. The color map jet

(Figure 11b) was preferred by the participants. We as-

sume that this color map preference indirectly draws

on experiential knowledge (weather representation,

from blue to red), leading to a corresponding asso-

ciation and increased intelligibility.

For the spatial binnings, the representation of user

activity in rectangles was selected by most of the re-

spondents (Figure 9a). This preference matches with

established representations. Similar rectangular grids

were already used by the pioneer of spatial obser-

vations William H. Whyte who investigated human

behavior in urban environments in the early 1970s.

Whyte was the ﬁrst researcher using a video camera

for long term spatial observations. His study proto-

cols are still leading information graphics in the ﬁeld

of urban studies (Whyte, 1980).

Regarding the applied colormaps a further simpli-

ﬁcation was evaluated within the user survey. This

approach consists of quantizing the continuous activ-

ity scale into coarser sections, as illustrated in Figure

12. The idea here was to clearly divide the activity

into referenceable blocks from “low” to “high”. How-

ever, the participants of the study clearly preferred the

full color gradient with 255 colors (compare to Fig-

ure 12d). Therefore, a division of the color scale has

not been considered further, since this also goes ahead

with a loss of information and makes it more difﬁcult

to identify hotspots on the plot.

5.3 Plausibility and Implemented

Changes

In the online survey, the plausibility was checked last.

The users were shown a map with three marked ar-

eas. They were asked to select in which marked area

the highest user activity was measured. 15 out of 16

people chose the correct answer. This conﬁrms the

comprehensibility of the presentation as well as the

presentation in the top view.

With the results from the survey, the visualiza-

tion was adjusted, and the sketch-drawing of the play-

ground was placed as an overlay on the rendering and

the user activity (compare with heat map in Figure

13). This allows the exact assignment of the user ac-

tivity.

6 CONCLUSION

In the conducted research we successfully explored

methods of collecting anonymized data that allow the

tracing of spatial use. The studied visualizations offer

a basis for the design of an integrated interface that

can be used by urban planners to evaluate the inter-

action within a determined area. The project aims to

Visualization of Activity Data from a Sensor-based Long-term Monitoring Study at a Playground

153

Figure 13: Example for prototyped user-interface.

collect open data in the terms of the Smart City to in-

form stakeholders and thus to provide an objective ba-

sis for discussion of decisions and changes. The com-

plex results are to be made available to urban planners

and interested citizens in an interactive interface.

This interface, an example is given in Figure 13,

will give the opportunity to select individual time

frames, spatial focus and further details of weather

conditions and other measurements that are easily

available and give insights about the circumstances

during the use of the site.

Outlook

In the current visualizations, only the measured ac-

tivity level of a single class (PERSON in the exam-

ples shown in this work) is considered and displayed.

Therefore, one aspect for further work is the evalu-

ation of methods for a consolidated visualization of

activity caused by a set of multiple classes. One goal

here would be to allow a further differentiation of the

underlying activity levels per class. This could be

achieved through the interactive usage of different vi-

sualization layers or the integration of spatial icons in

this activity visualization.

A second step in the further development would

be the improvement of the sensor signal processing

pipeline to achieve a continuous object tracking in-

stead of a semantic segmentation. This leads to the

question of how a long-term aggregated representa-

tion of these directed object movements should be vi-

sualized. These aspects should also be evaluated sub-

sequently in the form of another user-study.

ACKNOWLEDGEMENTS

We thank Felix Lentzen for his technical support and

work related to the described Blender based bird eye

view mapping.

Funding

This work was supported by the European Re-

gional Development Fund under grant number EFRE-

0801082 as part of the project “plsm” (https://plsm-

project.com/).

REFERENCES

Bolleter, J. (2017). Counter intelligence: evaluating wi-

ﬁ tracking data for augmenting conventional pub-

lic space–public life surveys. Australian Planner,

54(2):134–144.

Bolten, T., Lentzen, F., Pohle-Fr

ohlich, R., and T

onnies,

K. D. (2021a). Evaluation of deep learning based

IVAPP 2022 - 13th International Conference on Information Visualization Theory and Applications

154

3d-point-cloud processing techniques for semantic

segmentation of neuromorphic vision sensor event-

streams. arXiv.

Bolten, T., Pohle-Fr

ohlich, R., and T

onnies, K. D. (2021b).

Dvs-outlab: A neuromorphic event-based long time

monitoring dataset for real-world outdoor scenarios.

In Proceedings of the IEEE/CVF Conference on Com-

puter Vision and Pattern Recognition (CVPR) Work-

shops, pages 1348–1357.

Cuellar, K., Jambor, E., Sorornejad, P., and Talpey, S.

(2020). Improving visitor evaluation at the british mu-

seum. Worcester Polytechnic Institute.

Eghteabs, C., Ritter, B., Parikh, A., and Lischke, L. (2017).

Effects of color and threshold on user perception of

heat maps. In Proceedings of the 2017 ACM Con-

ference Companion Publication on Designing Inter-

active Systems, pages 275–279.

Gehl, J. and Svarre, B. (2013). How to Study Public Life.

Island Press, Washington, Covelo, London.

Geospatial, C. T. P. D. (2021). Use cases. https://cesium.

com/use-cases/. Online; accessed 2021-08-20.

Guo, M., Huang, J., and Chen, S. (2017). Live demonstra-

tion: A 768 x 640 pixels 200meps dynamic vision sen-

sor. 2017 IEEE International Symposium on Circuits

and Systems (ISCAS), page 1.

Hunter, J. D. (2007). Matplotlib: A 2d graphics environ-

ment. Computing in Science & Engineering, 9(3):90–

95.

Lauressens, T. (2005). Pedestrian levitation.net. http:

//www.visualcomplexity.com/vc/project.cfm?id=255.

Online; accessed 2021-08-20.

Moreland, K. (2009). Diverging color maps for scientiﬁc

visualization. In International Symposium on Visual

Computing, pages 92–103. Springer.

Nielsen, S. Z., Gade, R., Moeslund, T. B., and Skov-

Petersen, H. (2014). Measuring human movement pat-

terns and behaviors in public spaces. a method based

on thermal cameras, computer vision, and geographi-

cal information system technologies. Measuring Be-

havior 2014.

Peysakhovich, V. and Hurter, C. (2017). Scanpath visual-

ization and comparison using visual aggregation tech-

niques. Journal of Eye Movement Research, 10(5).

Qi, C. R., Yi, L., Su, H., and Guibas, L. J. (2017). Point-

net++: Deep hierarchical feature learning on point sets

in a metric space. In Proceedings of the 31st Interna-

tional Conference on Neural Information Processing

Systems, NIPS’17, pages 5105–5114, Red Hook, NY,

USA. Curran Associates Inc.

Rajasekaran, R., Rasool, F., Srivastava, S., Masih, J., and

Rajest, S. S. (2020). Heat maps for human group ac-

tivity in academic blocks. In Business Intelligence

for Enterprise Internet of Things, pages 241–251.

Springer.

Rashed, M. G., Suzuki, R., Yonezawa, T., Lam, A.,

Kobayashi, Y., and Kuno, Y. (2016). Tracking visi-

tors in a real museum for behavioral analysis. In 2016

Joint 8th International Conference on Soft Computing

and Intelligent Systems (SCIS) and 17th International

Symposium on Advanced Intelligent Systems (ISIS),

pages 80–85. IEEE.

Rezaei, M. and Azarmi, M. (2020). Deepsocial: Social dis-

tancing monitoring and infection risk assessment in

covid-19 pandemic. Applied Sciences, 10(21):7514.

Silva, S., Santos, B. S., and Madeira, J. (2011). Using color

in visualization: A survey. Computers & Graphics,

35(2):320–333.

Whyte, W. H. (1979). New york and tokyo: A study in

crowding. Real Estate Issues, 4(2):117.

Whyte, W. H. (1980). The social life of small urban spaces.

Conservation Foundation, Washington, D.C.

Zhou, L. and Hansen, C. D. (2015). A survey of colormaps

in visualization. IEEE transactions on visualization

and computer graphics, 22(8):2051–2069.

Visualization of Activity Data from a Sensor-based Long-term Monitoring Study at a Playground

155