Senseable City Lab :.:: Massachusetts Institute of Technology
Total Page:16
File Type:pdf, Size:1020Kb
Senseable City Lab :.:: Massachusetts Institute of Technology This paper might be a pre-copy-editing or a post-print author-produced .pdf of an article accepted for publication. For the definitive publisher-authenticated version, please refer directly to publishing house’s archive system SENSEABLE CITY LAB SMART CITIES Noninvasive Bluetooth Monitoring of Visitors’ Length of Stay at the Louvre The ubiquity of digital technologies is revolutionizing how researchers collect data about human behaviors. Here, the authors use anonymized longitudinal datasets collected from noninvasive Bluetooth sensors to analyze visitor behavior at the Louvre Museum. ecent emerging technologies— behaviors have trouble distinguishing be- along with their subsequent rapid tween individuals when the density of visitors diffusion into our daily lives— is high,4 and such cameras can’t always track have caused a structural change visitors as they move beyond the area of a par- in human behavior analysis. In ticular exhibit.1 Rparticular, the ubiquitous presence of wired To address these issues, we employed a Blue- and wireless sensors in contemporary urban tooth detection technique,5,6 which has many environments is producing advantages over other technologies. First of Yuji Yoshimura a detailed empirical record all, depending on the specifcation, Bluetooth Senseable City Laboratory, MIT of individual activities. Fur- can have a more fnely grained detection scale thermore, in addition to the than passive mobile phone tracking.7 Second, Anne Krebs ubiquity of sensors, computa- Bluetooth detection successfully works inside Louvre Museum tionally advanced computer buildings, where GPS connectivity is often lim- 8 Carlo Ratti systems make it possible to ited. Finally, in contrast to RFID or ultra-wide- 9 Senseable City Laboratory, MIT accumulate large datasets of band, previous participation isn’t necessary to human behavior at high fre- equip any devices or tags or to download the quencies—sometimes even in proper application in advance. Because prior real time. participation or registration isn’t needed, we can However, despite the widespread use of such perform data collection for longer than just one data collection technology, the analysis of visi- day or a few days, letting us generate large-scale tor behavior in art museums has not advanced datasets of human behaviors. much over the last few decades. The traditional All of these advantages make Bluetooth track- pencil-and-paper-based tracking method is ing the most adequate methodology for our still commonly used to time and track museum research. It doesn’t replace simulation-based visits,1 partly because many of the technologies analysis, which can estimate visitors’ move- that researchers started using a decade ago for ment in the museum,10 but because simulation human behavior data collection2 don’t work often requires a simplifcation of human be- properly in a museum setting.3 For example, haviors, Bluetooth tracking can complement video cameras installed to observe visitor the analysis by generating relevant datasets in 26 PERVASIVE computing Published by the IEEE CS n 1536-1268/17/$33.00 © 2017 IEEE a consistent way and letting us analyze For our study, eight Bluetooth sen- of stay in the museum. This analysis is real and large-scale empirical data. sors were deployed throughout the De- largely based on our previous research. non wing of the Louvre, covering key The second factor relates to entry times, Methodology: Bluetooth places to capture visitors’ behavior. which are used to assess the distribu- Tracking System Figure 1 presents the approximate loca- tion of visitors’ lengths of stay in the Bluetooth tracking systems for hu- tions of the sensors (nodes E, D, V, B, S, museum, depending on when they en- man behavior data collection can also G, C, P) and some of the most represen- tered the museum. The third factor pro- complement more traditional social tative artworks or areas of the museum. vides visitors’ lengths of stay near each sciences’ qualitative and quantitative The museum’s administrative policies specifc node, which correlates to a cer- methods. This is particularly useful for (such as those related to artwork pro- tain exhibit area, and the fourth factor museum visitor studies, which have a tection or areas with restricted access) is the relationship between the length of long tradition of employing qualitative and certain technical or spatial restric- stay at a specifc node and the number interviews, questionnaires, and ethno- tions (such as the circulation conditions of visitors around the node (density). graphic observations. (See Eilean Hop- inside the galleries) largely determined per-Greenhill’s “Studying Visitors”11 the installation locations and sometimes The Path Taken for a review of such studies.) prevented placement for optimal detec- Figure 2a shows that the median A Bluetooth tracking system works tion. As a result, some sensors are situ- length of stay is very similar across as follows: When a Bluetooth-acti- ated next to the relevant artworks, while all amounts of unique visited nodes vated mobile device enters the detect- others are located near but not right next (for example, if someone visits nodes able area, a sensor receives the emit- to the artworks (see the detailed spatial E-V-B-E, the number of unique visited ted signal from the mobile device until relationship in Figure 1). The detec- nodes is 3—E, V, and B). The differ- the signal disappears. Thus, the sensor tion range of a sensor can be 20 meters ence of visitors’ lengths of stay between registers the time at which the signal long and 7 meters wide and can be cus- all unique visited nodes and the mini- appears—referred to as the “check-in” tomized for smaller scales if necessary. mum one (visiting just two nodes) is less time. When the signal disappears, the Although a sensor’s range can fuctuate than one hour. In addition, the num- sensor records the “check-out” time. depending on its location, each one cov- ber of unique visited nodes seems to Then, the difference between each mo- ers the targeted areas. have a slightly negative slope with the bile device’s check-in and check-out We performed the data collection at lengths of stay until four unique nodes time can be calculated to defne the different periods using a different num- are visited. This means that individuals length of stay at the node (note that ber of sensors over fve months, from who visited three or four unique nodes “node” refers to the detectable area April through August 2010. All of this tended to rush during their visits, as op- formed by each sensor). Similarly, by information was collected without in- posed to individuals who visited one or looking at the frst check-in time and vading visitors’ privacy, because we two unique nodes. the last check-out time over all nodes, applied a secure hash algorithm encryp- To compute the correlation, we used it’s possible to calculate how long a tion to each sensor by converting each a nonparametric correlation analy- visitor stays in the study area. device’s media access control ID into a sis (Spearman’s rank correlation co- Such a series of check-in and check- unique identifer. effcient), because the variables don’t out time data, registered by installed After data cleanup and processing, in seem to follow a normal distribution. sensors, makes it possible to con- which we adjusted the data to remove We also include a series of boxplots to struct a human trajectory throughout any inconsistencies, we selected 80,693 better explain the relationship between the study area, including the travel unique devices to be analyzed for this variables. A very low correlation value time between nodes. In fact, because article. By comparing the number of de- (ρ = 0.073, p < 2.2e-16) suggests that Bluetooth detection offers system- tected mobile devices and ticket sales, there is no relationship between these atic observation through unobtrusive we found that, on average, 8.2 percent two variables. That is, the number of measures, a considerable amount of of visitors had activated Bluetooth on unique nodes visited seems to be almost research has employed this methodol- their mobile phones. independent of the length of stay. ogy for human movement tracking.5,6 On the other hand, Figure 2b shows More technical aspects of Bluetooth Factors Related to Length a different relationship between the detection—including information of Stay lengths of stay and the number of vis- about different Bluetooth anten- Here, we analyze four different factors ited nodes. The correlation coeffcient nas, the discovery time, and detec- related to visitors’ length of stay in the (ρ = 0.186, p-value < 2.2e-16) suggests tion interferences—are summarized museum. The frst factor deals with vis- a weak association between the two elsewhere.4 itor routes in relation with the length variables. The total number of visited APRIL–JUNE 2017 PERVASIVE computing 27 SMART CITIES Lower Richelieu –1 ground floor Entrance/ exit 1 E Sully 12 Entrance/exit Psyche and Cupid, A. Canova Denon 34 Captive (The Dying Slave), Righting Warrior, known as the Michelangelo “Borghese Gladiator” Connections lower ground and ground floor (Gallery Daru) 0 Ground floor 56 5 Aphrodite, known as the The Winged Victory of “Venus de Milo” Samothrace C 1 m P 10 32 4 78 D V5 Greate Gallery (Italian Mona Lisa, L. de Vinci paintings 13th–15th C.) Connections ground and 1st floor 9 10 1 1st floor Richelieu The Wounded Cuirassier, Great Sphinx of Tanis 10 T. Géricault Sully Denon S 6 1 m Location of a sensor (E, D, V, B, S, G, C, P) Artwork (1-10) 9 Areas unreachable to visitors 8 Estimation of sensor’s range G 7 B Figure 1.