5 min read

Video Redaction Using AI

Video Redaction Using AI
January 31, 2022
min read
share this article:

In recent years, artificial intelligence (AI), machine learning (ML) and deep learning (DL) have received increased significance in the technological sector. 

A multi-disciplinary subset of computer science, artificial intelligence (AI) is concerned with developing smart technology capable of performing and automating activities that would otherwise need the use of human intellect. 

This is especially true in recording, processing, and editing video content and information, where AI is used in various ways. 

To put it another way, AI would be all of the information and knowledge you already possess, whereas machine learning is associated with the methods you would use to acquire that knowledge, such as learning, perceiving, studying, or even making errors. Here is a great article here with a simple explanation of AI vs. ML vs. DL.

Understanding Artificial Intelligence

Artificial Intelligence (AI) is the capacity of machines, most notably computer systems, to perform tasks often associated with human intelligence. These mechanisms include comprehension (acquiring knowledge and rules for utilizing data), reasoning (applying rules to arrive at likely or decisive conclusions), and self-correction.

AI can be classed as either weak or strong. Weak AI is a term that refers to an artificial intelligence system that has been built and trained for a certain job; it is also referred to as narrow AI. 

Digital personal assistants, such as Apple's Siri and Amazon's Alexa, are a sort of weak AI. On the other side, strong AI refers to an artificial intelligence software that possesses extensive human cognitive abilities, also referred to as artificial general intelligence. 

When confronted with an unknown problem, a strong AI system should be able to find a solution without human intervention.

Artificial Intelligence and Video Redaction

Video is crucial in a variety of industries worldwide. For example, in law enforcement, video evidence enhances public safety by broadcasting live events, aids in creating trustworthy digital evidence, and even helps improve police conduct in some cases. 

Nonetheless, video processing and its use as evidence continue to provide enormous challenges, especially in terms of storage, security, and the anonymity of individuals, as law enforcement attempts to comply with the Freedom of Information Act (FOIA). 

What happens when body camera and dashboard camera footage is collected from a 500-person police department throughout an entire shift? And when one incident on one of those cameras is needed in a criminal case. How do state security organizations quickly and efficiently comply with subpoenas and evidence requirements while also meeting FOIA obligations and privacy restrictions? 

The primary issue is, police departments lack the time, people and money necessary to redact video recordings (obscure faces and car plates) manually in accordance with the law and prior to release. It is rather labor-intensive and costly, since it requires a team of people several hours to sift through and evaluate the recordings and manually cut portions of the video that may violate individuals' privacy. 

As body-worn and dashboard cameras become increasingly prevalent, more hours of video evidence must be analyzed and edited. This is where AI becomes advantageous since it can lower labor costs while still protecting citizens' privacy.

How does AI power the Sighthound Redactor software?

Sighthound's artificial intelligence and machine learning technologies are based on deep neural networks incorporated in the computer vision software we've created over the last eight years.

The AI system that drives Sighthound Redactor contains a state-of-the-art object detector that enables the software to discover individuals, faces, vehicles and license plates in a range of environments and circumstances. It is trained at detecting, tracking, and redacting any object in videos.

Artificial Intelligence is growing increasingly better than any human at recognizing when faces and number plates appear in the video especially with different camera angles, distances and lighting conditions; the AI inside Redactor has learned and been trained how to accurately and efficiently recognize objects and locate specific patterns and characteristics over the years. 

The Sighthound AI is an example of "Strong AI" or artificial general intelligence—it can analyze a video it has never seen before yet still detect and redact faces or plates without a human (in the mode termed automated redaction

AI-enabled redaction software is essentially like having a superhuman that can view videos at 5x the regular pace, detect all the sensitive material and blur it before a human could even watch the first minute of the video. 

It used to take a person many hours to watch one video over and over to find each face, even in the distance or in a reflection, draw a box and blur it. With Sighthounds AI-powered redaction, the computer learns to complete the work better and quicker, saving law enforcement organizations precious time. 

It also means that no technical skill is necessary to utilize the program. If we go back to the police agency example, any authorized redaction user can easily and quickly redact footage using the automated “AI” mode, yet still  review the footage and make any changes as they are the experts. 

After all, individuals are more knowledgeable than the machine regarding which may require redaction for any reason (for example, separating suspect from bystander or blurring out a minor).

AI ensures privacy

Pattern comparison is used by AI-based video editing applications to identify objects inside a video clip. The redaction software is pre-programmed to recognize individuals by their heads or the characters on license plates for example. 

Once a person has been recognized, the program operator can select which aspect of the footage should be hidden throughout. After redacting a specific object or subject, the finished output seems as if the object was never shot in the first place, with no accompanying identification. 

The AI user may then review the redacted video export to confirm that all essential adjustments have been performed entirely and precisely. Unlike complex systems requiring hours of training, AI-powered video redaction software is accurate and precise, sometimes achieving up to 98% accuracy in automated mode (before human review). 

With these systems in place, software users can ensure sensitive information is removed while ultimately encouraging accountability and building trust through the speed and efficiency with which privacy-protected video may be released to the public, individuals or regulators.

How Artificial Intelligence reduces costs

Rather than physically erasing information frame by frame, redaction software powered by AI identifies and provides a chronological list of potential video components of interest, ranging from license plates to human faces. 

Following that, users will choose what they want to redact, and the program automatically removes them all. A task that previously used to take hours of rigorous and costly effort, sifting through hours of video, can now be completed in a matter of minutes. 

Organizations, companies and agencies stand to save significant money with a more efficient, optimized redaction workflow. It is difficult to hire people in many countries around the world today with the “great labor shortage”, but also wages and “over time pay” are increasing, and employees need to do more and more tasks. By using a Redaction software with a strong, powerful AI capability, organizations can reduce the amount of time their employees need to sit and redact footage and can be redeployed to more productive and/or profitable activities.,

Save time and money with a simple, powerful tool!

Get started with Sighthound Redactor