Cameras produce huge amounts of data, most of which is completely uninteresting. The US Navy, like many other large institutions, feels this problem acutely because the large volume of video from the many cameras on ships fills up disks in a matter of hours. As a result, relevant information is either not recorded or lost in the noise. The Real-Time Evaluation, Compression, and Organization of Recorded Data using Extensible Rules (RECORDER) tool from Charles River Analytics, a GRVTY company, addresses this problem by making the recording process more customizable, efficient, and intelligent to maximize the value of the data saved. The Naval Sea Systems Command awarded Charles River a contract of up to $1.75M to develop and expand an associated prototype.
Selective motion-detecting methods, like those used by residential security systems, are impractical for the Navy’s purposes because of the ocean’s constant motion and the current approach of filling up disks only to throw them away is grossly inefficient. Recording at lower fidelity might cram more data into less space, but potential objects of interest would be rendered at lower resolution and not very useful. “What we really have is an optimization problem, determining what is getting recorded and saved, how to sort the value of the data that’s saved while minimizing the size,” says Ross Eaton, Director of Marine Systems at Charles River and Principal Investigator on RECORDER.
The RECORDER effort has three primary goals, Eaton says: “to create a way for users to specify the recordings that might be of interest, to share the reason why these things are interesting so we can understand what kinds of compression we can apply, and to use that information to maximize our usage of the available disk space.”
Understanding what needs to be recorded and what’s interesting about it helps inform RECORDER’s decisions about what kind of data—and how much of it—to keep. For example, if the goal of the recording is to retrieve information about ships near the horizon to classify them, we do not need recordings at the standard 30 frames per second.
Operators can specify the ontology or class of recordings of interest, such as “any ship that comes within 500 meters” or “whale tail flukes”. They can further tweak the search parameters by applying time and other constraints, like recording only between dusk and dawn. These targeted parameters sharpen RECORDER’s focus, ensuring it captures high-value data while filtering out unnecessary noise.
Data compression can take various forms, including spatial and temporal techniques, with bit rate compression specifically applied to video. These methods enable RECORDER to prioritize the most relevant data for the operator. “We’re not trying to eliminate the data. We’re increasing the amount of data that is interesting and increasing the available disk space by 90%,” Eaton says.
RECORDER can be integrated into future versions of Charles River’s Awarion® system to increase its appeal to government agencies and commercial interests, such as the offshore wind industry, as well as a host of other sectors that seek to maximize the value of their camera data.
Being able to have an extensible ontology that specifies the kinds of recording needed will help RECORDER find a number of commercial applications beyond the Navy, Eaton says. “We’re putting together a system where we can swap out the maritime ontology for a sports ontology or a speech ontology. Whatever you envision it to be, so you can build a system that is relevant in a variety of domains,” Eaton says.
RECORDER is now in Phase II development. “Scaling a prototype to include a large amount of video is an exciting opportunity to address a growing problem that many organizations are eager to solve”, Eaton says. “We’ve got a really interesting problem and a solution that we feel is going to be good. We’re introducing the concept of compressing videos into a format that’s easy to access and engage with, rather than letting them sit unused, metaphorically collecting dust on a disk,” he adds.
Contact us to learn more about RECORDER and our capabilities in computer vision and perception, and robotics and autonomy.
This material is based upon work supported by the Naval Sea Systems Command under Contract No. N00024-24-C-S169. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the Naval Sea Systems Command.
