Edge Computer Vision

Bring video analytics to new heights

Video-analytic market is expected to grow very fastly by end of 2020.​Atos wants to take this growth opportunity by combining its key assets (BullSequana edge servers, worldwide service capacities) with high-value added video-analytic partners.​

99% of enterprise-captured video/image content will be analyzed by machines rather than humans, ​ up from 30% in 2018

 

By 2022, more than 50% of enterprise-generated data will be created and processed outside the data center or cloud.

The solution

The Atos Edge Computer Vision is a complete environment running on BullSequana Edge and dedicated to deliver computer vision capabilities.

It provides advanced feature extraction (person, faces, emotion, behavior), feature sharing and classification, automatic actions based on feature analytics, real-time and post-event actions or post event analytics.

A powerful search function accelerates the search of a specific person from multiple criteria. It enables a large set of intelligent cameras to collaborate holistically in real-time, enabling operations to be tracked without interruption.

The Edge Computer Vision relies on the exceptional hardware acceleration for machine learning applications supported by Bullsequana Edge.

Thanks to its highly customizable architecture, Edge Computer Vision can be easily utilized in many various use cases such as homeland security, the tracking of vehicles, the crowd movements, the detection of abandoned objects, or to analyze a population density or expression and improve the quality of service in shopping or advertising.

Video surveillance

Traditional video surveillance

Traditional video surveillance encounter constraints, as human workers have a limited attention span. Indeed, research suggests that they suffer from fatigue after only 18 minutes, leading to poor monitoring performance and anomalous events being missed. Also, a security officer is not able to check more than 5-10 cameras over a 5-minute period in crowded areas (CPNI Recommendations). Without any assistance, a single tracking of person of interest performed by human workers requires in average 3 security officers. Hence, video analytics solutions are flourishing to improve situational awareness to ensure safety and security.

People efficient tracking is hard:

  • Need real-time processing​
  • People from back, from far​
  • No cross camera FoVcontinuity​
  • Keep people privacy​

Edge Computer Vision

For years, some rule-based solutions have been used to analyze images. With the advent of artificial intelligence and deep learning, computer vision can now analyze images in real time. Atos solutions allows body recognition from different angle of view to track a person, ​

Atos developed a complete solution to help cities to take immediate actions when unexpected event occurs. This solution offers real-time tracking and live people search tool to assist security staff. The tracking can be started from any people detected on the camera network. To spot a specific person within the camera network, a multi-patended body recognition feature is used. The reidentification occurs on different angle of view of the person : front, back, side.

Security officers can take immediate actions from abnormal situation or behavior. They can automatically start the tracking on the person of interest and take appropriate action to solve the problem.
They can also search in live stream for lost person (kid for eg) and drastically reduce associated traditional search’s cost. Hence, it allows cities to enforce GDPR and respect people privacy by not using Facial recognition for video stream analysis.

Benefits

Take immediate actions​

​Increase security officer’s efficiency in tracking and searching tasks​

​Reduce cost of monitoring and lost person search.​

​Respect people privacy: No Facial recognition​

Deploy centrally or at the edge​

Why deep learning in video surveillance?

  • Major breakthroughs in computer vision with deep learning and GPU in last decade
  • Now, the machine can equal or surpass the human-level in video analysis tasks with pretty good results on any real-world data
  • Emerging deep learning extensions to improve further the robustness in worst conditions

Strong expertise in deep learning

  • 10+ patents in DL-based computer vision and tracking
  • 5+ publications in international conferences
  • Experts team in deep learning model design & optimization
  • Industrial & academic partnerships

Use cases

People safety & security

  • Eldery people
  • Missing kid
  • Violence detection & police support

Retail

  • Store analytics
  • Autonomous shopping

Smart cities

  • Transport safety
  • Traffic management

Remote surveillance

  • Data center
  • Oil & Gas filling stations

Manufacturing

  • Quality control

Related resources

Our experts

Portrait of expert

Mehdi Kasmi

Global Head of business development – Bullsequana Edge

Portrait of expert

Matthieu Schmit

Global Head of presales – BullSequana Edge

Interested in our video analytics solutions?