Content-aware bandwidth adaptation on edge AI devices

Use-case : Driven by builtin AI detection at the edge, the content-aware bandwidth adaptation allows to spare the bandwidth with critical-data only

As live showcased at the renowned Viva Technology 2025 event held in Paris from June 11 to June 13, this innovative feature provides the capability to conserve bandwidth effectively, thanks to the advanced AI detections processed directly at the camera side. The content is meticulously filtered in real-time under live conditions by our built-in AI, which can also be enhanced with specific regions of interest and designated regions of exclusion. This sophisticated approach facilitates the precise setup of essential parameters, including Frames Per Second (FPS), Bandwidth (measured in Kb/sec or Mb/sec), and Resolution (expressed in pixels), enabling output only when critical data is detected within the image or within the defined regions of interest.

Furthermore, multiple classes can be utilized, such as person, car, truck, and many others. Additionally, custom classes can be designed and implemented in bespoke projects to suit unique requirements.

In the video below, you will observe that humans are detected within the clearly marked rectangular region of interest, which is highlighted in a vivid red color. If there are no humans detected within this specified area, the video stream is generated at a reduced rate of 1 frame per second, with a bitrate of 500Kb/sec and a resolution of 640x360 pixels. It's important to note that only critical data, specifically the instances of humans crossing within the designated Region Of Interest, is streamed out, maintaining a nominal stream quality consistent with that of the original camera feed.

Spare up to 50 times less bandwidth

--

Spare up to 50 times less bandwidth --

All the parameters are addressable via the PAI Pod (framerate, bandwidth, resolution, class of object, region of interest).

This feature will be available in the PAI Pod starting in Q3 2025, as a standard software feature.
This is compatible with any camera stream and with any VMS, making networks more efficient and cost-effective.

 

Useful for : Telco, Banks, 4G/5G networks usage, Smart Cities, Archives, …

Next
Next

Visual access control : Yellow Vest and Security Helmet Detection