File size is a major factor adding to complexity – while a 500-page document is around 20 MB, that amount of data is only about 20 seconds of video. With camera technology becoming more sophisticated all the time – wider panoramas, higher resolution, and capabilities like compression, streaming, storage, and analytics – file sizes will just keep increasing, straining abilities to manipulate, send and store files. 

A second complicating factor is the proliferating web of state and local laws governing the reasons for recording video in the first place and then managing the data ongoing in the public records context. With officer-involved shootings front and center in the public eye day after day, and multiple sources of video evidence, jurisdictions are spending enormous amounts of time struggling to provide relevant legislative frameworks and managing the release of footage accordingly. It’s not getting easier or clearer anytime soon.

Another formidable hurdle with audio and video files, unlike other digital files, is that they take special skills and crippling investments of time to redact to meet burgeoning legal requirements, including privacy protection – measures like blurring or blacking out heads, license plates, and home interiors.

A welcome development is the ability of AI-enabled software to automatically and efficiently identify and redact portions of images that must be obscured to protect privacy – providing a key solution to the complexity of video/audio record requests. 

GovQA has partnered with Veritone to integrate video and audio redaction tools into the GovQA workflows, making redaction up to 90% faster than other methods. Whether using sophisticated AI tools or manual methods of redaction, however, the more video files an agency must process for public records requests, the more complexity increases.

“Today, BWCs are likely the most rapidly diffusing technologies in modern police history….“

Body‐worn cameras’ effects on police officers and citizen behavior: A systematic review, September 2020

The most recent available national statistics on use of body-worn cameras (BWCs) in policing come from the U.S. Bureau of Justice Statistics from 2016 data. At that time, 60% of local police departments and 49% of sheriff’s offices had fully deployed BWCs  – double the number in use in 2013. 

With agencies stating the main obstacle to implementing BWCs as the cost of hardware, video storage, and system maintenance, from 2016 to 2020 Congress appropriated $112.5 million for a grant program to help law enforcement agencies onboard BWC systems. 

Just one real-life story, though, puts that amount into sobering perspective. As reported by the News-Gazette covering central Illinois in relation to the City of Elgin’s consideration of mandating bodycams by 2025: 

“The Elgin police department already implemented body cameras in 2015, which costs the department $170,000 each year in data storage for its force of 182 officers, according to Elgin Police Chief Ana Lalley…[Chief] Lalley told lawmakers the city would not have been able to implement the program if it did not receive a $500,000 federal grant.”

So that’s almost half of one percent of the entire federal grant fund for a single city of 100,000. A number of jurisdictions, including Portland, OR, have cut police bodycam pilot programs due to cost/budget issues – and Oregon jurisdictions can charge requesters for redaction time, unlike many others.

The Washington Post reported in 2019 that “In recent years, after a spate of fatal police-involved shootings sparked nationwide protests, politicians and community activists seized on police body cameras as a way to restore public trust. But although the cameras were widely adopted, many departments — especially in smaller jurisdictions — are now dropping or delaying their programs, finding it too expensive to store and manage the thousands of hours of footage”- including East Dundee, IL (pop. 3,000, which would have spent $20,000 per year on bodycams).

Not only are BWCs proliferating, law enforcement agencies are creating and gathering a multitude of other types of video and audio not only to aid in their work but in response to public demand for transparency and accountability in the face of high-profile incidents involving police use of force. These include dashcams, publicly and privately collected surveillance video, audio and video recordings of interviews, footage from drones and helicopters, imaging from automatic license plate readers (ALPRs), and a host of others. 

In 2020, the American Bar Association reported:

“In 2018, California became only the third state in the nation [following South Carolina and Nevada] to require that all police departments in the state deploy BWCs.  Notably, after the killing of George Floyd, four more states—Colorado, Connecticut, New Mexico, and New York—have mandated that all police departments deploy BWCs. As a result, there are now many more BWC recordings of police-civilian interactions than six years ago, dramatically increasing the stock of BWC recordings that are potentially available for public viewing.”

According to the National Conference of State Legislatures (NCSL), as of April 2021, seven states have mandated the statewide use of body-worn cameras by law enforcement officers: Colorado, Connecticut, Illinois, Maryland, New Jersey, New Mexico and South Carolina. While the ABA article includes New York and Nevada, neither New York’s nor Nevada’s legislation appear to be blanket mandates for all law enforcement agencies in the state – patchwork situations with the potential to create even more complexity.

The ABA’s characterization of of CA’s BWC mandate as applying to all jurisdictions appears to be mistaken. California does not mandate BWCs statewide. So why the confusion?

State (and local) laws relating to law enforcement video (especially bodycams) are becoming increasingly tangled webs governing what agencies have the ability or obligation to record video and under what circumstances, who has the authority to release it to the public (if at all), when the release occurs, the length of time recordings must be retained, what has to be redacted and by whom, parties that do and don’t have standing to request footage, camera capability requirements, and on and on. According to the National Conference of State Legislatures (NCSL), less than half of U.S. states have legislated how body-worn camera data is addressed under open record laws.

Media reports of the aftermaths of officer-involved shootings are focusing more and more on the release of video footage. Take the roughly coincident shootings of Andrew Brown in Elizabeth City, NC and of Adam Toledo in Chicago in spring 2021. On the Brown shooting, the media reported: 

“A judge has ordered a delay on the public release of bodycam footage in the case of Andrew Brown Jr., the man shot and killed by deputies last Wednesday in Elizabeth City, denying a request from the media to make the video immediately available to the public.

“The release at this time would create a serious threat to the fair, impartial and orderly administration of justice,” said Judge Jeffrey Foster, [who] said the video won’t be released for a minimum of 30 days (or a maximum of 45 days) as authorities investigate.

“However, all footage from multiple body cameras will be made available to be viewed by Brown’s immediate family (with redactions allowed) within 10 days. Brown’s family has seen one 20-second clip from one bodycam so far.

“Once the 30 to 45-day window is reached, the judge will determine if the family can obtain a copy of the video, which they could then publicly disseminate.

“The judge also ordered all faces of officers involved in the incident to be blurred before footage is released.”

On the Adam Toledo shooting, the Chicago Civilian Office of Police Accountability (COPA) issued a press release that included the following:

“Today [COPA] informed the Toledo family….that video and other materials related to the fatal officer-involved shooting of 13-year old Adam Toledo will be released tomorrow, Thursday, April 15, 2021.

“COPA has remained sensitive to the family’s grief and is carrying out this release in accordance with the City’s Video Release Policy. COPA’s core values of integrity and transparency are essential to building public trust, particularly in incidents related to an officer involved shooting, and we are unwavering in our commitment to uphold these values.

“This incident which occurred just 16 days ago, and the materials released will include relevant body-worn footage, 3rd party video, OEMC transmissions, ShotSpotter recordings, Case Incident, Tactical Response and Arrest Reports…Tomorrow’s release is in accordance with the City’s Video Release Policy and well in advance of the 60-day deadline.”

The circumstances and tone around the release of police video in these two incidents seem starkly contrasting. Even if media coverage makes some mention of state laws playing a role, most people are largely unaware that the release of video is handled so differently from case to case because state and local laws differ so vastly (in the Andrew Brown case, North Carolina does not consider police video to be public record at all). To the public, unexplained differences can contribute to the feeling that governments act arbitrarily and seek to obfuscate the truth – fuelling anger and mistrust.

The average per-officer BWC produces 32 files, seven hours and 20GB of video per month.  In California, that translates to 550 law enforcement agencies averaging 100 officers each equipped with BWCs – amounting to over 38,000 files, almost 8,500 hours of video, and over 23 terabytes of data per year – a massive amount. 

While video and audio collected by law enforcement agencies is usually considered public record, it is subject to a host of exemptions including those relating to individuals’ privacy, confidentiality around certain law enforcement practices, integrity of ongoing investigations, and other factors.

So just as with paper or electronic documents, exempt data has to be redacted from video and audio, and an exemption log created to document what was redacted. Over the time that video public records have started being collected and requested, this redaction has proven to be incredibly painstaking and time-consuming – on the order of five to ten hours to redact one hour of video, even with traditional video editing or BWC-paired software offered by camera manufacturers to leverage efforts.

From the technology perspective, you could say that the problem of having to spend hours redacting video and audio comes down to the difference between structured and unstructured data.

But before we explain, an example from California. As we detailed here, in 2020 the state’s Supreme Court decided a case involving whether a city police department could charge a requester for time spent to redact video footage from officers’ response to a demonstration (they were providing mutual aid to another city, which adds complexity to the cost issue). The city initially identified 141 video files totaling 90 hours that were potentially responsive. 

The city asked the requester, who agreed, to narrow the scope, reducing the request to six hours of video. Even with the significant reduction and help of software, the redaction took 40 hours. The city billed the requester for around $3,000, which the Supreme Court ultimately concluded state law did not authorize. If the requester had not agreed to reduce the scope, the bill would have come to around $45,000 – representing what is really an unsustainable use of staff and public funds.

Software can recognize, search for, and redact digitized documents very quickly – even if the words and letters don’t have identical appearances (different fonts, etc.). The reason? This data is structured – it is easily recognizable and “regular” software can readily read it. 

Video and photography images – and specific elements within them such as faces – are much more complex. They require our eyes’ and brains’ cooperation to recognize a multiplicity of forms and contexts – lighting, angle, color, partially obscured, in motion, etc. This kind of data is unstructured – not in an easily recognizable code.  The process for recognizing that something is a face or a license plate instead of a street sign, for example, takes intelligence.

This blog post by the Seattle Police Department shows how far things have come in just a few years – in 2015 the city sponsored a hackathon to develop processes to redact video from its pilot bodycam program: 

“Using [a new] process, it only took half a day to redact more than four hours of footage…The old method would have required days of work. A simple manual redaction in a one minute video, for example, can take specialists upwards of half-an-hour, whereas more complicated edits—like blurring multiple faces or pieces of audio—can take much, much longer.”

You can probably see why SPD’s objective at the time to immediately release all bodycam video online to a Youtube channel didn’t pan out.

And this is where artificial intelligence comes in.  We all know the term, but its meaning and applications aren’t always clear. Computers and software have been able to recognize and manipulate structured data for quite awhile, but as this source puts it:

“When you give [cameras and computers]  a file, whether it’s an image or a video, understanding isn’t automatic.  To machines, a picture or video of a license plate isn’t any different from one that shows a football player or a skyscraper or a sweet potato.”

But AI builds “intelligence” into software so it can now do much more complex tasks like recognize and redact faces in widely varying contexts. 

GovQA has created an available integration with Veritone digital evidence management software to create an end-to-end solution for responding to requests for video and audio that greatly reduces redaction time – from five to ten hours for redaction of one hour of video – down to one hour of redaction time. 

The realm of law enforcement video is transforming rapidly on all fronts, making it increasingly difficult for jurisdictions to put a pin in the map and decide if, and how, they will implement camera programs. With officer-involved shootings and reports of alleged misconduct coming thick and fast, there is greater demand for accountability and transparency. Research and study are providing more new input all the time on whether video tools like bodycams in fact serve their intended purposes – or whether benefits may be illusory. Technology is transforming with lightning speed, and vendors with gee-whiz innovations market directly to law enforcement agencies under pressure from all sides. Lawmakers struggle to keep up with appropriately regulating potentially explosive new information sources and sophisticated tools.  And finally, the true costs of video programs – not in initial acquisition but long-term costs to store, manipulate, and provide footage to the public is becoming starkly clearer. 

The solution to the challenge of reviewing and redacting increasing stores of  video and audio files in order to provide them as public records will be seen in the PiPR Index data over time, as more agencies adopt AI redaction tools and allow them to start getting costs under control as a result. We expect that video file counts will continue to increase; but time to process will start to move inversely.