The goal of this project was to take a corpus of videos and transcripts from police body cameras of civilian encounters with police officers, tag them with various metadata about the encounter, and then use that data to try to find trends in what actions resulted in positive and negative outcomes in order to improve police training.

I worked with the researchers on this project to create a model of the metadata that would need to be collected and design the CMS interface to collect the data and a search interface. I also built the interface in html and Spring to integrate with a Java search engine built by someone else on the project.

Initial Research

The research team was distributed among several different research institutions, and all of them turned out to be fairly concrete thinkers who had trouble thinking about the metadata they would need in the abstract. I solved this problem by setting up an online meeting with screen sharing and used my wireframing tool to take notes. That helped me turn the question from "what types of information do you have?" to "how do you enter it?" which jogged loose a lot more information about special cases, additional related data, etc.

Below is a cropped version of my notes from one of the meetings. You can click through to see a larger image with even more boxes and arrows.

Prototyping

After getting some notes for each part of the system, the notes were broken up into a multi-page interactive wireframe, which I shared with the researchers and tested with some of the real data to make sure that how things were connected together made sense to the people who would be entering the data, that the UI flowed well with the form the data was given to them to prevent skipping back and forth (and subsequent errors and omissions), and that it was possible to enter all of the necessary data correctly. (The wireframes were in PDF form, so it wasn't possible to actually test data entry, but we worked through it aloud.)

Most of the screens underwent several revisions in this phase.

Implementation

Once revisions started getting down into wording details I put together an html prototype and prepared for implementation.

We ended up creating an entirely custom CMS with the Spring framework because the engineer working on the search engine portion was a Java programmer and I have yet to encounter a Java CMS that really accommodates fully custom content types well, but if I had it to do over again I would have insisted on a framework in some other language (my current framework of choice for such things is Django) that could call a Java program; I believe we would have gotten much more consistent, maintainable results that way.

Below is an animation of adding encounter and participant data for a video. It's fairly dense in the number of fields, but they did need to be able to record and search all of those things. The key point was getting a layout that made it easy to scan all of the items and easy to add multiple participants per encounter and multiple encounters per video.

Stop animation