How accurate is Secure Redact?
The technology
Pimloc uses computer vision and machine learning to train its Secure Redact product to recognise personal data in images and video.
Secure Redact recognises the identifiers (faces, heads and vehicle registration plates) as a human observer would, so it can redact and anonymise PII automatically.
The technology can be deployed selectively, so it can remove some or all faces, and/or all vehicle registration details, or other identifying information. And it also works on low-definition materials, meaning Secure Redact can deal with the diversity of CCTV and other video surveillance data at scale.
As a result, video clips that have been selectively anonymised with Secure Redact’s technology can be supplied externally as needed - for example to employees, customers, or even corporate insurance companies, law enforcement and lawyers, all while staying compliant with data privacy laws and regulations.
The challenge
An hour of CCTV can contain 2 million faces. Manually redacting every image of personal data from every frame is an expensive, boring and difficult task that requires considerable attention to detail and can take users days or even weeks to complete.
With Pimloc’s AI, this task can be completed 200x faster - enabling companies to be far more efficient and accurate when it comes to anonymising video footage, and opening up new opportunities for its use.
Secure Redact’s service is critically acclaimed for how accurately and quickly its machine learning platform can automatically detect personal identifiers, in order to redact them.
System evaluation
We are confident in our technology, so much so that we offer potential customers a free trial to see the results of our solutions themselves - they just have to sign up on the Secure Redact website.
We have painstakingly built our own test data to measure our system performance against real-world footage, ensuring world-leading performance for PII anonymisation in real-world CCTV, body-worn and dash cam footage.
We set our internal target accuracy levels much higher than is required by our customers and legislation as we want to provide tools that can blanket protect personal data in all videos.
We have a team dedicated to continually improving our system performance for the benefit of our customers and the public's privacy. We have designed our artificial intelligence solutions from the ground up specifically for video privacy, and that’s allowed us to closely integrate all areas of our product - allowing a tight feedback loop for ongoing improvements, and resulting in world-class system performance.
We have many customers across a range of sectors who vouch for the product in critical Security, Education, Transport, Retail, Insurance and Healthcare.
Why Secure Redact is different?
Pimloc has developed models that provide accurate detection and redaction in the most challenging circumstances. This means creating deep learning algorithms that have been trained on faces, heads and number plates across a domain-specific video from security and road survey video footage.
It is these algorithms, our end-user applications and APIs that now provide our unique advantage.
We have accurate solutions that excel at detecting and redacting personal data within low-quality, variable CCTV footage - that no other solution is currently able to handle at scale.
System bias
Our machine learning experts constantly work to update and tune the Secure Redact platform, so we are always working towards consistently bettering our performance. Our technical approach to PII detection mitigates the usual bias challenges in traditional face detection: because we are not analysing faces, our system is less susceptible to negative biases in ethnicity, age and other demographics. We are building systems that can protect the personal data of everyone on the planet, regardless of how they look or how they were captured.
Our approach to PII detection is also less sensitive to facial occlusion and change than normal systems trained to recognise faces - this means the system performs equally well for people with facial hair, masks, glasses and other facial artifacts.
Secure Redact accuracy
Given the diversity of CCTV, Body-worn and dash cam footage, no solution will ever be 100% accurate across all real-world video content, but that is something we are striving towards.
The Secure Redact system provides maximum recall rates currently exceeding 99% of identifiable PII in security video.
Customer fine tuning
Pimloc can work closely with strategic partners to further improve system performance based on customer data. Pimloc can tune the system based on example frames from customer videos and/or can recreate similar video frames based on an understanding of the customers’ video domain.
Pimloc is also working on an update to Secure Redact that will allow premium users to further improve their system performance by annotating example frames from their video directly, which will provide enhanced performance for the customer on their data.