Background
Museums within any society reflect social and cultural changes on how the country has progressed over time. The future role of museums is constantly changing, and museums find themselves in a dynamic environment where traditional methods are constantly being challenged. To survive, thrive and compete with other attractions, museums need to be more adaptable to meet the needs of their visitors, while not losing sight of their mission. Furthermore, with the onset of the Covid-19 pandemic, museums across the world have experienced sharp declines in visitors in 2020. For example, The Louvre, the world’s biggest museum, suffered a drop in visitor numbers of over 70 percent in 2020.
It is critical for museums to embrace new methods to collect more actionable data, gaining deeper insights on art piece specific performances, visitor demographics, and analyse visitor behaviours to continue attracting visitors. Having these data points will help museum management to:
1. Better curate its collection of art pieces to suit the intended target audience
2. Restructure the museum layout to optimize traffic and improve the overall visitor journey
What are we trying to solve?
Are existing metrices/analysis sufficient?
Ticket Sales
Existing performance assessments are largely based on annual ticket sales or the number of visitors. These measures only offer high level information but fail to address the crucial question of -
“Which art pieces are visitors interested in and what can be done to attract more visitors?”
Traditional Data (Surveys/Polls)
Opinions on cultural interests or feedback towards specific exhibits and art pieces are sometimes gathered through customer surveys, which face the perennial issues of low response rates, non-response bias and time inefficiencies.
Current metrics used are broad and do not allow museum management to derive many actionable insights. One key limitation of existing analyses will be the lack of granular data on visitors at the art piece level. Most museums currently do not track the:
- Total number of visitors that have viewed each art piece
- Duration taken for visitors to view each art piece
- Profiles of interested visitors by art piece
Having these metrics on top of current ones can help museum management determine the popular pieces, i.e., pieces that have attracted more visitors and have captured visitors’ attention for a longer time. Museum management can better understand consumer preferences with this data (e.g., interactive vs non-interactive art pieces, art pieces by a specific artist, or thematic art pieces) and adjust their collections accordingly. Other areas that warrant further investigation include visitor movement throughout the exhibits, to glean insights on visitor behaviour and aid in decisions on the optimal placement of art pieces. Data on visitor demographics also allow museums to understand the target audience for each exhibit or art piece.
Introduction of Smart Museum Suite (SMS)
To allow museums to collect the aforementioned data and derive actionable insights, our team came up with the Smart Museum Suite (SMS). The target stakeholders would be the museum management, museum operations-level employees and visitors.
SMS is a combination of:
- Beacons (i.e., wearables) worn by visitors, with dual purposes: as devices used for exhibit admissions and for sending signals to sensors that detect real-time visitor locations.
2. Receivers (i.e., sensors) installed within exhibits to detect signals from nearby wearables, monitoring the number of visitors at each art piece, and the duration spent by each visitor at each art piece.
Data will then be collected and processed, before being visualised in a dashboard, where insights and recommendations can be obtained.
High-Level Solution Design
MVP Architecture
This is a high-level view of our MVP architecture implementation, using micro:bits as both our wearables/beacons (1) and receivers (2). The receivers are installed near art pieces and are connected to laptops, which also act as gateway nodes, creating a one-hop star-tiered topology. Upon receiving data packets from the beacons, the receivers/gateway nodes process and store data in-memory, only to publish it to ThingSpeak, an IoT analytics platform service, periodically (3). From ThingSpeak, we download our data in an Excel file, perform further processing (4), and finally, feed the data to Tableau for visualisation and analysis (5).
Data Collection and Processing
The collected data enables end-users to determine critical information such as the closest art piece to a visitor at any point of time, as well as the total duration spent at an art piece for each visitor. However, multiple processing steps are required to achieve this due to the limitations of message publication in the free version of ThingSpeak, which only allows 8 data fields and restricts publication to 1 message in 15 second intervals.
Data collection and processing at receivers
The count of packets received is required to determine the art piece/receiver that is closest to a visitor at a point of time, in scenarios where more than 1 receiver picks up the same packet from the same beacon (explained further in the next paragraph). Using the highest RSSI to determine the closest receiver is unreliable due to fluctuations in micro:bits’ RSSI values.
Data processing on data downloaded from ThingSpeak
Data processing at this stage aims to achieve 2 objectives:
1. To assign the receiver with the highest count of data packets received from a particular beacon as the nearest receiver. E.g., If receivers A and B picked up 10 and 3 packets from beacon 1 respectively, beacon 1 will be assigned to receiver A.
2. Calculate the duration spent at each receiver/art piece for each beacon/visitor.
At Scale Architecture
When deployed at scale, the micro:bit wearables are replaced with BLE beacons (1) instead as it is more cost efficient, and will also incorporate the demographic information of the visitor (please refer to the At Scale — Visitor Demographics section for more details).
Instead of each receiver being tied to only one art piece, multiple receivers will be deployed throughout the exhibit (2) to triangulate the location of each wearable, allowing for better scalability. A meshed network topology is used for higher reliability, as data can be routed through multiple receivers in the event that any receiver fails.
Subsequently, receivers send the data to gateway nodes that are located in each exhibit, before the data is sent via MQTT (3), a messaging protocol for device-to-device communication, to a cloud server such as an Amazon Elastic Compute Cloud (Amazon EC2) (4). This can then connect to a relational database such as a MySQL server, using the Amazon Relational Database Service (Amazon RDS). Thereafter, Alteryx, a tool that specialises in Extract, Transform, Load (ETL) functions can be used to build an end-to-end data pipeline that cleans, blends, and prepares data to be visualised.
Finally, a live connection to the Tableau dashboard can be refreshed periodically, not only allowing for quicker insights to be gleaned by museum curators and management, but also allowing operations-level employees to be alerted via mobile in the event that faulty devices are detected.
From Data to Wisdom
Collecting the Raw Data (D of DIKW)
Using a similar process described under High-Level Solution Design, the SMS captures sensor data in a format that contains the date, timestamp, wearable ID and signal strength received (for the at scale solution) by each receiver.
Deriving the Information (I of DIKW)
Information such as the art piece that is being viewed by a visitor based on the wearable ID, on which day, and at what time can be derived.
Making Sense of the Data (K of DIKW)
The SMS dashboard is customised primarily for management to draw interesting insights and uncover trends which will help them better curate their collection of art pieces, assess the placement of art pieces in exhibits and optimise current crowd management measures.
Dashboard 1 : Tracking Visitor Flow
Calendar View
A high number of visitors was observed on 2nd April 2021, which was a public holiday (Good Friday). The targeted stakeholders will also be able to get a sense of which art piece is more likely to attract the visitors’ attention by viewing the chart in the tooltip.
Visitor Movement across time of the day
Stakeholders will also be able to monitor the visitor movement across the day and identify the peak hours with the highest visitor footfall, as highlighted in dark purple. In the month of April 2021, the peak timing was found to be between 2pm and 3pm. In addition, the stakeholders can easily slice the data to view the distribution of the visitor movement by weekends/weekdays, or public/school holidays.
Dashboard 2: Distribution of Time Spent by Visitors
The trendline, as indicated by the blue line (median time spent by visitors on each day), helps to inform the stakeholders on how the average time spent on the art pieces vary across the different days. A general downward trend is observed from March to early April. They can perform the analysis by selecting all art pieces to get an overview, or drill down to the art piece level by selecting the art piece filter. Management can use this information to identify the art pieces that are able to capture the visitors’ attention more.
Dashboard 3: Visitor Demographics (At Scale)
The various visualisations displayed on this dashboard allow management to further understand the visitor profiles at each exhibit or interested visitors of each art piece. In this example, visitors of Exhibit A comprise a good mix of males and females and most of the visitors belong to the age group of 40 to 49 years old (29%). In addition, most of the visitors visit the museum either in groups (35%) or alone (35%).
Dashboard 4: Popularity of Exhibits and Art Pieces (At Scale)
The layouts of each exhibit and the heat map, as shown above, indicates the number of interactions between visitors and the art pieces, which are derived based on the triangulation of the exact locations of each wearable. The number of interactions is encoded in the colour scheme, which ranges from dark blue to bright red. In Exhibit C, art pieces which are placed in the corner of the room were found unattractive since they require visitors to walk through an additional route. This likely explains why the area does not show up at all on the heatmap and is represented by the red circle.
Dashboard 5: Overall Customer Journey (At Scale)
The visualisation of the customer journey path taken by the visitors from the moment they enter the exhibit allows museum management to understand which are the most common pathways and the corresponding art pieces that visitors are more attracted to. For example, as shown in the visualisation above boxed in red, visitors usually take this specific path — Art piece 7 -> Art piece 6 -> Art piece 9. It is also observed that the visitors are attracted to Art piece 7 as soon as they walk into the exhibit, bypassing the other nearby art pieces.
Dashboard 6: Health Status of Devices (At Scale)
This view can only be accessed by the operations-level employees, who will also have access to detailed information highlighting the exhibits of the defective receivers and the duration in which the wearables lost signal. Any wearable that has lost signal for more than one hour is flagged and identified as defective.
Benefits of SMS (W of DIKW)
Based on the insights from the dashboards shown earlier, the museum management will be able to implement the following actionable plans.
Learning & Discussion Points
Challenges & Limitations
1. Steep Learning Curve — As IoT is very new to all of us, we had to figure our way through the technical parts when working on our solution. The in-class lab sessions gave us a good introduction to using micro:bits and consultation/email queries to Professor Tan Hwee Xian and Instructor Pius Lee helped us immensely in formulating our eventual solution.
2. Limited Budget and Timeline — Due to the project budget and tight timeline, we could not explore other sensor types which could potentially help us to collect more accurate proximity data as compared to using micro:bits.
3. Deployment of Solution — As our target audience are museum curators and the museum management, we were not able to test out our solution in an actual museum setting. The testing and data collection were performed in a simulated environment instead.
4. Pre-processing of data — The raw data extracted from ThingSpeak requires significant data preparation before it can be used for our visualisation process. This was accomplished with the use of VBA Excel macros to calculate the duration that each visitor spent at each art piece.
Key Takeaways/ Final Thoughts
Overall, our team realises that while the Internet is a very prevalent part of our lives, not all problems require IoT as a solution. Coming up with a coherent and concise problem statement for our project was very challenging, but at the same time, it was vital in ensuring that we did not stray away from addressing the essential issues that we initially identified.
Furthermore, we learnt that testing our prototype in real life proved to be more challenging than what was expected, and we had to be adaptable along the way to change our set-ups to better meet the needs of our project. With regards to the more technical aspects of our project, we had to carry out more research about the different networks and available sensors before deciding on what was most appropriate for our project.
As we look to the future of Smart Museums, IoT will play an integral role alongside more sophisticated technologies such as facial recognition, in order to offer museum management with increasingly greater levels of actionable insights.
The team behind this amazing idea!
We would like to thank Prof. Hwee Xian and Pius for guiding us throughout the project design and implementation.
