A privacy-forward urban sensing project.
Team: Jae Kim | Alan Ren | George Verghese
Sensing in the Public Domain
Sensing has become an important and rapidly demanded method of understanding our built environment. Be it from understanding the energy performance at a building scale to using footpaths on the street. These demands come from increasing efficiency, optimizing performance, and creating a path forward backed by empirical evidence. This brings to question the policies, methods, and goals of sensing in today’s urban environment.
A lot of what we call “tech” today lies within device-level applications. However, urban tech operates in a grey and currently undefined area. The device layer is the city, and the applications are the deployments of sensors within the device. However, no individual user within the device can make decisions on data governance; it lies either with the city or with the urban tech company’s altruism. We must then demand and understand how we think of privacy policies in the public domain. Is it a generic privacy agreement each citizen signs, or perhaps at the sensor level with every possible attempt to preserve privacy made?
Our team chose to test what it would require to develop and design a sensor with privacy in mind. Setting out to solve this problem, we took an interest in understanding what activities and flows in our public space are like. This process required understanding how much data is needed, our end-use cases, and potential sensors.
Through an iterative process, we realized that privacy through software alone was not entirely perfect, and it required a hardware level of intervention. We aimed to implement a thermal sensor built with an edge compute node that can quickly process and out binary data without ever requiring raw data storage. Using existing computer vision algorithms to create human tracks, we set out to understand and enumerate our public spaces within Avery Hall. Through experimentation and analysis, we tested and learnt of the costs and benefits of using such hardware in our public domain.
“sensing in the public domain operates in a grey area, where participants are unaware of the data governance and policies, nor have the agency to elect out of the data collection”
the team
Sensor Build
Our sensor suite is built using off-shelf development boards and enthusiast parts. These are not commercial-grade equipment and have their limitations.
- NVIDIA Jetson Nano 2GB
- Raspberry Pi v2
- FLIR Lepton 3.0
- EdiMax WiFi USB
Data Collection & Analysis
The initial utilization of the thermal camera was to underscore the significance of preserving the privacy of individuals. Thermal cameras, by design, capture heat signatures emitted from a person’s body rather than capturing their facial features or other identifying characteristics. The rationale for using this technology was to monitor and analyze human movement and behavior while minimizing the risk of privacy invasion
However, during our study, we encountered unforeseen privacy concerns when the thermal camera inadvertently captured the outline of individuals’ undergarments. Consequently, this issue led to the reconsideration of using thermal cameras for this research, and the primary reliance shifted to webcam technology. Despite the unintended privacy implications associated with thermal cameras in this particular study, the initial motivation for using them highlights the importance of carefully considering the privacy of individuals in research projects. Balancing the need for accurate data and maintaining the confidentiality of the subjects is a crucial aspect of responsible research and should be prioritized throughout the research process.
Figure 1 portrays the aggregate human movement traces over a 20-minute period on the first floor of Avery Hall, while Figure 2 displays the same data for the entrance of Avery Hall. The primary distinction between these two figures is the recording perspective: human eye level for Figure 1 and an elevated view for Figure 2, allowing for a more comprehensive understanding of the entrance area.
The information gleaned from Figure 1 is limited due to the human eye level perspective. The paths are predominantly consistent, originating from or leading to the stairs. A notable observation is the presence of two individuals seated in chairs on the right side of Figure 1. However, the overlapping lines make it difficult to discern specific human behavior patterns from this visualization.
Learn more about the project at: https://gv2325.github.io/heatbodies/ or through the project report. Our code and project files can be forked from the link above!