Solving For Today's Video Analytics Challenges

Many aspects of life are utilizing video analytics to solve problems and make business and people’s lives smarter.  Whether this is a Smart building where secure points can be controlled via facial recognition for far better security than a badge, to monitoring COVID-19 requirements for face masks and social distancing, to Smart City applications that monitor traffic congestion and detect crime.

Hundreds of millions of cameras are being deployed in the cities, retail stores, railway stations, manufacturing lines. The eyes are there, but is the brain keeping up? In fact, the ability to extract insights from this overwhelming quantity of information has been more challenging than ever. Companies that help develop and deploy video analytics solutions for Smart Building and Smart City management are faced with one or more of these key challenges:

  • Massive amounts of data are collected from cameras passively, but because of the lack of computing power at the edge the recordings are erased before they can get analyzed.
  • Streaming data from hundreds of edge cameras 24 hours a day to high-powered servers in the data centers or the cloud to extract insights would incur millions of dollars in bandwidth cost.
  • Processing the incoming video. It requires significant adaptability to manage a bewildering variety of different video sources and types of encoding, to say nothing of compute power required for video frame extractions, getting everything into the correct color space, and scaling it to the proper size for the AI model to work on. 
  • How to enable intelligence on existing cameras right away without excessive cost and hassle, and without changing the existing camera setup.
  • Limited availability of robust Real-time Video Analytics platforms that offer high accurate AI models for edge computing.

Aupera Video AI Analytics for Critical Infrastructure Applications

Aupera is a key Xilinx partner who are able to effectively address these challenges. They’re a company that provides highly intelligent video processing solutions from the cloud to the edge. They’ve focused on providing highly efficient and agile deployment-ready AI applications on Xilinx Zynq® UltraScale+™ MPSoCs and Alveo accelerator platforms. Aupera’s video analytics solutions transform all passive camera data into actionable intelligence while achieving TCO saving, power efficiency, and bandwidth savings. Most importantly reducing deployment complexity. 

High Performance, Low Latency Video Analytics Platform

AI models and deep learning are key technologies that are used to gain insight on video enabled applications. Getting insights accurately on these videos is compute intensive and complex, often requiring multiple neural networks to run concurrently to achieve high accuracy with deterministic latency. With streaming increasing exponentially and an ever-growing number of cameras deployed, use of generic CPUs which perform all processing in software has become a critical bottleneck. To remove the CPU bottleneck in video processing, Aupera has innovated a whole new distributed micro-node architecture based on Xilinx Zynq® UltraScale+™ MPSoCs, to build an adaptable computing platform for video transcoding and real-time analytics.

Zynq MPSoCs have hardened video codecs capable of low latency simultaneous encode and decode, up to 4K resolution at 60 frames per second. The programmable logic on the Zynq devices provides the flexibility to run different video AI algorithms in parallel efficiently to deliver accurate results at a deterministic low latency, resulting in industry-leading low Total Cost of Ownership (TCO) versus GPU-based video analytic solutions.

Adaptable Framework for Edge, Cloud and Data Center Deployment

Aupera provides a simple deployment model that enables everything from connecting to multiple cameras to intelligence output with both pre-built and customizable neural network models. Aupera’s AI video solution includes a full software stack that encompasses video gateway, ready-to-deploy AI models for inference acceleration, and a set of standard APIs for easy integration with an enterprise application or 3rd party software platform.

Aupera has created ready-to-deploy video analytics appliances that help solve the lack of edge computing power and deliver the ability to extract valuable at-the-edge insights by moving computing closer to the cameras. Aupera appliances enable higher performance, lower latency and reduced bandwidth expenditures for the customer.

The Aupera V205 Edge portable appliance supports 8 channels@1080p30 video AI, and the Aupera 2601 Edge server supports 64 channels@1080p30 video AI. For data center and cloud deployments customers can use the Boston Stream AI Appliance, a 2RU server powered by Alveo U30 data center accelerators that supports 112 channels @1080p30 video AI.

aupera graph.ai

Best-in-Class Accuracy AI Models

Creating AI models from scratch can be time consuming and expensive for developers building Smart City and Smart Building applications. Aupera has created several production-ready, highly accurate pre-trained models to support diverse Video AI workloads, including facial recognition, crowd statistics, people tracking, virtual fencing, car tracking, car license plate recognition, and video anomaly detection.

For developers who would like their own custom AI model, Aupera provides a seamless integration path to mainstream frameworks like Caffe, Pytorch and Tensorflow for model development and deployment.

Want to learn more? Join us for a webinar where attendees will get insights on Video AI analytics market trends and how Aupera solutions are deployed across multiple industries in Smart City, Smart Retail, Smart Building and Smart Healthcare. Register at Real-Time AI Video Analytics for Smart Cities (xilinx.com)

For more info about Aupera’s Real-time AI Video Analytics solution: https://www.xilinx.com/products/acceleration-solutions/aupera.html