Umbo — Building Better Computer Vision For Better Security
Challenge
Customers in the security industry have long had to deal with the challenge of monitoring and responding to the 7.2 billion hours of security video footage recorded every day. Human operators are expensive, find the job exhausting and are frequently ineffective. Some form of automation in the monitoring and alerting of threats in security footage would save organizations money, free up security managers to do their jobs, and better protect people from harm.
Intelligent Video Surveillance (“IVS”) systems were developed to address this need but fall far short of the task. They require tedious setup configuration for each scene they observe and trigger a tsunami of false alarms at obviously harmless things such as snow, reflections, shadows or blinking Christmas lights. For this reason, operators are quickly frustrated with the “intelligence” of these IVS systems and ignore their alerts.
Developers at Umbo Computer Vision set out to overcome these challenges in creating a better autonomous video security system with Umbo Light.
Solution
Umbo’s engineers realized that Light needed to do better than just simple motion-tracking which is what older IVS systems did. Motion-tracking gives little context about what is actually happening — it could be a car driving by or two people fighting. Security managers would want to be alerted on the latter. To achieve this, Umbo’s system needed to “see” the contours of the human shape at the pixel level. No currently available model can do this so it had to be created.
Umbo Computer Vision’s research division developed custom computer vision models powered by the cloud. To train its models, Light used images from Umbo’s one-million-plus live video sessions streaming across 50 countries. And to further improve performance, each Umbo SmartDome camera creates over time its own individual model tailored to the scene it is pointed at.
This approach has yielded drastic improvements that directly translated into cost and labor savings with intrusions are alerted on with over 97% accuracy. Since its Sept 2016 launch, Umbo Light has processed over 700 million images, replaced over 160,000 operator-hours spent monitoring footage, and saved customers over $1.6 million in operator costs.
“Pixel-level segmentation was the key to helping computers see the same way humans do.” commented Business Development Manager Edward Chen. “Doing this in the cloud helped keep our models on the cutting edge and at their most effective.”
Impact
With Umbo, security managers are now able to harness unique computer vision models that allow them to monitor large perimeters with just their video cameras. This allowed them to save money while providing better protection to their clients.
For more information, visit www.umbocv.com
