Automated video redaction
The growing picture
At a conservative estimate there are 1.25M CCTV cameras in the UK and 30M across the United States. These generate over 4 billion hours of video every week, capturing each of us around 70 times a day.
And these numbers are growing. Between better affordability and new capture formats such as body-worn, vehicle-mounted and drone cameras, more and more organisations are finding how valuable video can be to their business.
Surveillance is critical and increasingly prevalent to public safety and efficient policing.
Autonomous vehicles capture thousands of faces, number plates and other personal data in their training operations.
Every business is storing video from around their offices, at events, trade shows, demonstrations and meetings.
The question you have to ask is: how easy is it for you to identify and remove the faces, number plates and other private data from these materials when you have to?
It might be an access request from a member of the public, or it might be requested for evidence, annotation or review.
Furthermore, you must consider “selective anonymisation”, that is the removal of all personal data except for specific individuals. This is incredibly time intensive to achieve manually, and is another example of a specific but valuable need.
The challenge is compounded by the quality of this data. Consider CCTV format issues, busy scenes, difficult angles, interference from weather, lighting and other technical difficulties.
GDPR and Data Protection
Governments around the world are quickly reconsidering the public’s right to privacy in digital content.
In Europe, GDPR now gives people rights to request, access and delete their own personal data.
This must be provided free of charge and data controllers have one month to respond.
This covers images of faces, unique body markings, number plates, house numbers and other piece of visual data that could be used to identify an individual.
The United States currently manages data protection law by industry, on a sector-by-sector basis, including laws and regulations developed at both the federal and state levels. These may be enforced by federal and state authorities, and many provide individuals with a private right to bring lawsuits against organisations they believe are violating the law. The State of California provides the best model for what is expected to come nationally.
“‘personal data’ means any information relating to an identified or identifiable natural person (‘data subject’); an identifiable natural person is one who can be identified, directly or indirectly, in particular by reference to an identifier such as a name, an identification number, location data, an online identifier or to one or more factors specific to the physical, physiological, genetic, mental, economic, cultural or social identity of that natural person.”
The Solution:
Automated protection for all, automation through deep learning
Historically, businesses have tried to automate redaction using ‘knowledge-based’ algorithms, which require developers to manually codify descriptions of faces and objects (ie. describe the lines of a face, the existence of eyes and a mouth).
However, these solutions perform poorly when exposed to real life video, due to the wide variance in camera angles, chaotic movement and occlusion of people within scenes and varying environmental conditions. It’s too complex to manually codify an accurate description of a face across these variables.
More recently, developments in a field of AI called ‘deep learning’ have provided new techniques for the fast and accurate detection of personal data with more flexibility. Deep learning systems are trained using millions of images so they can automatically detect and classify people, objects and scenes within image and video data.
But again, the general face and object detectors from the leading online AI cloud service providers perform poorly across the majority of security and road survey footage as they’ve been trained using lifestyle and celebrity style imagery. They’re just not used to seeing faces from strange angles in grainy footage within busy scenes.
We believe there’s a better way.
Through Pimloc’s specific work in this area, we have been able to tune models that provide accurate detection and redaction in the most challenging circumstances. This means creating deep learning algorithms that have been trained on faces and number plates across domain specific video from security and road survey video footage.
It is these algorithms that now provide the unique advantage of Secure Redact.
Secure Redact has been developed by Pimloc, world experts in deep learning technology, to provide automated redaction services for bulk video in the security and surveying markets. It allows businesses to protect people’s privacy whilst maintaining their current operations.
On premise vs cloud vs managed services
The sheer scale of current video collection requires many organisations to run their processing within an on-premise environment to avoid the latency and cost of uploading/downloading large files into and from the cloud.
Many security focused organisations have additional data privacy policies that mandate all their video has to be stored within their own environments and cannot be uploaded online. They have to run on-premise solutions with special provisions made for upgrades and support.
On-premise provision requires a central processing server with licensed users able to manage redaction jobs directly from their securely networked terminals/laptops. The specification of processing power requirements is dictated by the required number of concurrent users, volume of video being redacted and the criticality of redaction speed. Hardware requirements can range from a small standalone server costing a few hundred dollars to larger multi-GPU machines for larger teams/workloads.
Cloud services are a convenient solution for those with smaller and less frequent demand as upload/download times are shorter and the usage based cloud pricing models are less prohibitive for lower volumes.
Some organisations are not currently set up to manage the workflows and processes associated with bulk video redaction but need them to maintain critical on-going operations. In these circumstances, fully managed service provision is required to provide an end to end solution for all redaction needs. These services are provided to cater for more complex redaction and file management tasks.
For more information about the Secure Redact solution and to discuss your needs for on-premise, managed services or cloud based video redaction please contact: info@secureredact.co.uk