Commercializing computer vision research

As part of the MHCI capstone project, we were asked to help the Future Interfaces Group productize the research they had conducted on Zensors, a novel computer vision solution.

Team size

5

Roles

Design, research, code

Duration

2018, 6 months

Commercializing academic research

Zensors is the result of years of research conducted by the Future Interfaces Group, a research lab out of the Human Computer Interaction Institute at Carnegie Mellon University. The technology leverages the power of crowd computing, computer vision, and machine learning to provide readily accessible tools for collecting and analyzing visual data. Researchers involved with Zensors saw potential in developing it as a product that would democratize computer vision, making it easier than ever before to develop a deeper understanding of a physical space.

Our team was tasked with determining the most compelling product market fit for Zensors, focusing our design and development efforts on creating an valuable experience for users within a particular market vertical to optimize for adoption. From there, we would need to design and build a interface to layer on top of the existing Zensors infrastructure, paying close attention to providing the best possible experience for our target users.

Are there any seats available?

Y
E
S

How many cars are parked here?

2

How many products are left in stock?

2
6

What we built for Zensors

Over the course of 6 months, we conducted market research and identified a key target demographic for initial launch. After that, we conducted in-depth user research to guide the development of a functional MVP (with validated interaction flows and wireframes) that we connected to the Zensors back-end. In later stages we developed a scalable visual design system, and packaged up our continued user research as a resource for the rest of the Zensors team.

Scalable design system

We spent a lot of time validating core interactions, and making sure users had a sound understanding of the Zensors system. Once the validated flows were in place and implemented through our functional MVP, we layered on top a scalable visual design system that allows the Zensors team to build out new features building on top of our learnings.

1
2
3
4
5

Secondary navigation

As we investigated user mental models of the system, we eventually found that users managing coworking spaces think of location as a top-level entity, and try to find sub-entities within that layer.

The high level navigation allows the user to switch between locations (a natural grouping, as we found from our user tests), whereas the side menu functions as a tabbing interface between core components of the application. Using cognitive task analysis to study user mental models, we found that dashboard, data, sensors, and cameras made sense.

Handling natural language inputs

One of Zensors' unique features is its ability to handle natural language questions. This was challenging from a design perspective, because user expectations around machine learning are largely unexplored.

To help understand how users perceive the relation between cameras, questions, and sensors, I built a simple node editor that users could experiment with during testing. This allowed us to discover how potential users approach the question construction process, and how they think about the entities involved.

Eventually we found that such a freeform approach to natural language input and question creation led to confusion for the user and inconsistent inputs that the back-end couldn't handle. As a result, we decided to provide more structured inputs, with input hints along the way.

Expert tools for expert users

As users create value for themselves and add more sensors, locations, and camera devices, dashboards can become unwieldly to manage, and require complex navigational hierarchies. When not in use, the quick action has the affordance of a text input, and serves as a navigation indicator by showing the current page the user is on,. or the action the user is performing.

Visualizing complex datastreams

The main value of Zensors is derived from the ability to quickly glean valuable business insights from sensor data, not from consuming text content (where a lighter UI would be more sensible). The visual design of the application should support that goal.

Graphs showing Zensors data can be highly complex, and show multiple (overlapping) data streams simultaneously. High-contrast colors on dark backgrounds help users tell elements apart a lot easier than white backgrounds. Graphs are decorated with high contrast vibrant colors, keeping labels white for maximum contrast.

Modular data visualization

Since Zensors serves such a wide array of industries, the structure of data visualizations may differ from customer to customer. We designed modular data visualizations that consist out of core components such as zoomable timelines, graph types, and visualization composiiton elements to allow customers to built visualizations that suit their needs.

The graphs are built out of separate components, making it easy for the team to reuse and built new features on top of it. Besides that, users may also be able to customize their own graphs in the future.

Branding

Starting with a blank canvas, Zensors had little design assets to start with. As efforts to commercialize were just getting underway, I spent some time designing a logo that incorporated Zensors' key brand values, while simultaneously being a play on the functionalities and name of the platform. The mark incorporates an inverted wifi symbol, the letter Z, a stylized scanning eye, and an abstract silhouette of a monk.

Mapping out customer journeys

After identifying the coworking domain, we conducted site visits, contextual inquiries, and (remote) interviews to build a model of user behaviours.

We identified three categories, and built customer journey maps for all of them. Building these customer journey maps allowed us to identify key touchpoints of user interaction with the Zensors system, and highlighted points of attention for our design efforts.

On-site visits

As we worked to unpack the roles, processes, and goals of coworking we developed personas and experiences maps to synthesize key features and taskflows for the platform. These design artifacts were crucial in helping us identifying where to focus our upcoming design and development sprints in the summer.

Additionally, we built connections with a local Pittsburgh coworking space. We interviewed the operations director and community managers to understand their roles, responsibilities, and their data needs. The pilot partnership would allow us to deploy a Zensors solution in one of the coworking locations, and learn as the technology is being used.

Where we're leaving Zensors

Zensors has recently raised it's first round of venture capital and is now actively acquiring customers in government, coworking and many other industries. The presented designs are being implemented as needs arise and features are being shown to investors.

This page is a work in progress and will be updated as the project wraps up