Megh has developed the Video Analytics Solution (VAS) to solve the problem of inventory loss in the retail supply chain. The solution is targeted for various use cases including fraud prevention in retail locations, inventory tracking in manufacturing, and video surveillance for physical security. The solution runs on Megh's Real-Time Analytics Platform that maps the entire real time analytics pipeline consisting of the ingestion phase of streaming data input, transformation phase of video decoding and image resizing and the inference phase of object detection and classification into multiple networked FPGAs with integration into user's application. The Solution consists of Sira Accelerator Function Units (AFUs) for inline processing of streaming data and offload processing of compute intensive algorithms on multiple networked FPGAs. The AFUs used in this solution include:
• PPE (Packet Processing Engine) for ingesting the data using a direct NIC and extracting the payload
• SPE (Stream Processing Engine) for transforming the data which includes multi-channel H.264 decoder and image re-sizing
• DLE (Deep Leaning Engine) for image classification using CNN topologies like Resent50.
These are managed by the Arka Runtime which exposes Accelerator Functions-as-a-Service via high level APIs for integration with the applications frameworks like Spark, TensorFlow, kdb+, with no changes.