OStream’s PipeRunner is an edge AI media platform for converting video into an event-driven dataset. This solution provides real-time pattern detection to enable automated IoT workflows across applications from smart agriculture to smart cities. PipeRunner uses AI to detect and track context to accelerate media decoding while embedding useful insights into the video. OStream’s design can be clustered to form a scalable and cost-effective edge AI hardware platform using the following board.
Built by OKdo, OStream’s PipeRunner-RKAI uses a Rockchip 1808 SoC that features 3 TOPS AI processing at full frame rate. At the heart of this device is a dual-core Arm Cortex-A35 64-bit processor supported by an NPU and 1 GB of RAM for running Debian 10 LTS. This 96 x 76 mm board is additionally supported by 8 GB eMMC storage, Wi-Fi, 10/100 MB PoE, and Bluetooth 4.0 connectivity, and 1x 4-lane MIPI camera, 1x USB 3.0, and 1x USB 2.0 high-speed interfaces. PipeRunner-RKAI can be powered via USB-C power supply (5 V @ 3 A) and has an operating temperature from 0°C to +85°C.
OStream’s PipeRunner boards come preloaded with Physico software to make graphical AI pipeline building easy. They also leverage a pre-integrated set of AI computer vision stacks, GStreamer, and OpenCV to automatically run your elements on the right hardware accelerators for the task. In conjunction with Physico, this board detects and tracks context within your MP4 video stream to enable triggering of real-time action with other devices and systems, while simultaneously embedding useful metadata into your video stream.
OStream’s ready-to-go platform streamlines real-time pattern detection and simplifies development of automated AIoT workflows. It also enables users to search their media streams in a similar way to Google Images™ as objects are recognised and categorised.
You can learn more about PipeRunner-RKAI in this datasheet, but if you’re already wanting to evaluate this technology for use in a commercial application, apply for a sample using the form below.