Technology Overview
A Unique Fusion of Deep Learning and Computer Vision Technologies
Powerful & Easy to Use Video Analytics
At the core of BriefCam is a highly refined video content analytics engine developed by a team that’s headed by one of the world’s leading computer vision and deep learning experts. As we process video, we recognize and extract objects, along with information about the type and attributes of those objects, e.g. people, bags, vehicles, animals, size, color, speed, path, direction, dwell time, and more. The ability to extract, isolate and differentiate between independent objects, especially in scenes containing small or distant objects, poor illumination, background distractions or high activity, is very difficult. In order to accomplish this, our R&D team has solved hundreds of computer vision and deep learning challenges.
Overcoming Technology Barriers
Below are six major technology barriers we overcame to realize our vision of a powerful, flexible and easy to use solution.
Filtering Out Noise
To limit the false detection of new objects in a frame, we’ve learned how to filter out the movements in the environment – such as branches, shadows, reflections, waves and clouds.
Detecting Subtle and Camouflaged Objects
We can identify very small subtle objects, down to the level of small creature. While identifying an object in a frame that is appearing against a background of the same color is a difficult task, over the years, our detection mechanism has become very sensitive to subtle differences in color and texture.
Handling Complex Lighting Conditions On the Fly
Every scene is different, and over the years we have developed the capacity to accurately handle thousands of scenes with different lighting and weather conditions. You would be surprised how snow and rain impacts the accuracy of computer vision!
Understanding What Constitutes a Single Incident
If a person “disappears” from the frame for a period of several seconds, as they pass behind a car or a tree for example, we detect that, and treat it as one incident. If the person goes in and out of the frame boundaries, we treat it as separate object incidents.
Fusing of Deep Learning and Computer Vision
We have fused the best of both technologies to bring Smart Synopsis® to our customers allowing them to query for “women wearing black and white, riding a bicycle, travelling south at 10mph” and to display all results that might have occurred over a period of an entire day in literally seconds!
Integrating 3rd Party Metadata
Our ability to connect video time stamps to metadata from other systems to create tight custom integrations enables solutions like pairing an ID badge system with video information to video. The same goes with audio feeds, motion sensors, smoke detectors and more. Not only do you have the alert from the third party feed, but the video of what happened at that exact moment, along with more efficient presentation and rich query formulation capabilities.The Result?
A powerful and easy to use video analytics solution that enables customers to review hours of video within minutes, rapidly pinpoint people and objects of interest, receive real-time notifications of critical events, and dynamically analyze key performance indicators to optimize operations.