The concept of video redaction is simple, right? You have a face, license plate or some other personally identifiable piece of information that needs to be blurred out before being shared with media or the public. While simple in concept, developing a solution for this task is actually pretty complicated. I’ll take you through the journey of creating our video redaction software, Ikena Spotlight, what we learned along the way, and where we’re taking the software in the near future.
Where We Started
We released Ikena Spotlight over 5 years ago to fulfill a requirement from police in the UK to protect civilians’ identities in security camera footage. With an estimated 5.9 million closed-circuit television cameras, or one for every 11 people, you can imagine how challenging it is to work with that much video evidence.
Due to the massive amount of video collected and the requirement for timely public access to this video while adhering to strict privacy laws, police needed a quick and easy way to blur out faces of individuals prior to release. At the suggestion of the London Metropolitan Police, we decided to develop this software as a part of our Ikena Forensic software lineup.
What We’ve Learned
1. Speed is a crucial factor to consider.
There are a number of factors that affect the speed at which you can process video evidence:
- Videos from body cameras, in-car systems and security cameras are often quite long, sometimes hours, and these files are large and time-consuming to process. This is why MotionDSP uses GPU-acceleration whenever possible, as GPUs can process video orders of magnitude faster than standard CPUs.
- Cloud-based solutions are at a disadvantage to locally-installed software for few reasons. Since the file sizes are large (more than 1 gigabyte per hour), they take hours to upload and download. Next, due to delays caused by the internet, cloud-based systems respond slowly to the user’s input. Finally, they use shared servers, making computation slow (see bullet 1 above).
- Blurring a video requires changing the actual pixels in the video. This is computationally intensive (decode, display, blur, re-encode).
- Automated redaction without human interaction may sound great in theory as a time saver, but it presents a number of issues.
- Automating redaction using object tracking has been made further complicated by body cameras. It is relatively easy to track moving objects from a wall-mounted security camera that doesn’t move, but things get far more complicated when the camera itself is moving as you’d find when a police officer is running. This is where we have applied our tracking knowledge from our drone software. While there may be some redaction solutions that have a tracking capability, do they actually hold up to challenging scenes?
2. Redaction is only one piece of the solution.
Blurring out a face or license plate is the main requirement for redaction software, but there are other capabilities that are needed by law enforcement to prepare video that is suitable for public release:
- Audio redaction – needed to protect identities of innocent civilians and undercover police officers.
- Annotation – a scene may be so complex that you need to point out things with text to explain what is happening.
- Compatibility – redaction software needs to work with any video file type from any camera source. We often see software, such as video evidence management systems, that only work with the vendor’s own cameras and files types. You may be able to redact body cam video, but not cell phone, in-car, or security camera footage.
- Highlighting – rather than obscuring things, police officers may want to draw attention to a particular person or object in a video.
3. Developing good software is hard.
More than 15 years of university R&D went into our image processing algorithms before we even spun it out. Since then, we’ve spent another 11 years developing and refining it and adapting it for new customer requirements such as:
- New, more challenging video sources: body cams and in-car video systems where the camera “and” the redacted object are both moving
- New technology. NVIDIA’s release of CUDA, allowing graphics card to be used for computation, was a game changer for processing power and just in time as cameras moved to HD (1080P) and higher (4K) resolutions.
- New operating systems (Windows XP, Vista, 7, 8, 10)
Where We’re Going
We will release a new Ikena Spotlight this Spring. It will address the many new challenges that police departments face when handling video redaction, particularly as a result of the adoption of body cameras. The new features and upgrades we’ll be adding are the result of direct input from law enforcement professionals, and will give them the tools necessary to quickly redact video for public release and address privacy concerns.
As police adopt new technologies in the future, such as drones, a different set of privacy concerns and challenges will arise. Just as we’ve continued to update our software over time to work with all types of video formats and sources, we’ll continue adapting it to stay up with the latest technologies.