Amazon Kinesis Video Streams (KVS) is a video ingestion and storage service that allows users to stream videos to Amazon Web Services (AWS) for analytics, machine learning and other processing. AWS partners with companies specialized in video analytics or artificial intelligence solutions to enable users to build custom applications on the KVS platform and take advantage of these features. Since Agent Vi, Veritone and ABEJA are among the first companies AWS has partnered with, we asked them to talk about the benefits of such partnership and what they expect to see going forward.Here is how Amazon Kinesis Video Streams (KVS) works. First, KVS provides software development kits to manufacturers to install on their security cameras or other devices — this makes it easier for users to securely stream videos or any time-encoded data from their connected devices to AWS. Then, KVS ingests, stores and processes video streams at any scale for real-time and batch analytics, allowing users to quickly access and retrieve video fragments based on timestamps. Finally, KVS enables users to further analyze video data using video recognition or other applications that take advantage of computer vision and video analytics. It is during this last step that the strengths of AWS partners such as Agent Video Intelligence (Agent Vi), Veritone and ABEJA come into play.
The biggest advantage of KVS is that it creates and maintains a ready-to-use video ingestion system, in which all the infrastructure necessary to manage video streams is already in place. Users do not need to worry about configuration, scalability, available storage and data security as the number of streams grows; instead, they can focus on building their own artificial intelligence (AI) or machine learning (ML) applications to analyze video data.
KVS can be applied to many different scenarios. For example, in a smart home setting the user can stream videos from a baby monitor to AWS for simple playback on a smartphone or some more advanced tasks such as facial recognition. In a retail store where there are multiple security cameras installed, the owner could stream footage captured in real time to AWS and then analyze live feeds using video analytics applications to understand consumer behavior.
While AWS offers its scalable and cost-effective KVS service to end users, it has also established the AWS Partner Network (APN) program. The program is designed to help customers identify companies with ML competency and expertise to support integration and deployment of various solutions. The designation recognizes members who provide solutions that help organizations solve their data challenges and enhance machine learning applications. Both Agent Vi and Veritone are APN partners.
According to John Ward, VP of Marketing at Veritone, a company must first meet three requirements to become an APN partner. First, it must have more than four AWS customer references regarding completed ML projects. The references must be for projects that started within the past 12 months and are actually in the production stage. Second, its products or solution must be available in three or more AWS regions. Third, there must be a reference for an ML case that is optimized for security, reliability, performance, cost optimization and operational excellence.
innoVi, Agent Vi’s cloud-based video analytics software as a service (SaaS), is implemented by deep learning technology and other advanced algorithms that enable detection accuracy. Able to distinguish between people, vehicles and static objects, innoVi can not only detect security incidents in real time but also perform analysis that transforms security cameras into intelligent devices, uncovering previously unavailable information.
Veritone has developed aiWARE, an AI operating system that leverages AI-based cognitive computing, including facial recognition, transcription, geolocation and sentiment detection to analyze unstructured audio and video data, such as TV broadcasts and surveillance footage.
In a press release, Chad Steelberg, Chairman and CEO of Veritone said, “Institutions and organizations recognize the necessity of analyzing unstructured data at scale in near real-time. aiWARE provides them a way to unlock this data, accessing deep analytics and providing business insights like never before. Our collaboration with Amazon Web Services allows customers to deploy our platform in the cloud within minutes, giving them the ability to harness AI to make decisions with more confidence.”
The partnership between AWS and the aforementioned three companies presents a win-win situation. For Agent Vi, the integration of innoVi with KVS allows the camera owner to add smart analytics functionalities that can automatically detect and alert to events of interest. “The collaboration between Agent Vi and AWS brings a truly disruptive service to the market, that leverages advanced cloud and AI technologies to make any camera, of any brand and from any location, seamlessly smart within seconds,” said Itsik Kattan, CEO of Agent Vi, in a press release.
One of the biggest benefits brought by the integration is that it saves hardware costs for customers. “Access to the surveillance footage is obviously mandatory to employ innoVi, and we have so far implemented it using our own innoVi Edge, which is an appliance deployed in the customer's network that serves as a cloud-gateway between the cameras and our cloud platform. Now, using the integration with Kinesis Video Streams, the video will be streamed directly from the cameras to the AWS cloud, allowing us to access the video in the cloud directly, for further analysis. This simplifies the customer's deployment, eliminating the need and cost of the innoVi Edge,” Kattan said.
Kattan added, “Now that the integration with Kinesis Video Streams can eliminate the need for innoVi Edge in certain cases, innoVi can become a pure SaaS solution. This opens up opportunities regarding how to make innoVi more available to potential customers.”
With the integration of Veritone aiWARE and KVS, users are free from the trouble of managing a huge amount of video data while enjoying the advantage of aiWARE making every frame of video searchable for objects, faces, keywords and more.
“Amazon Kinesis Video Streams can benefit Veritone by allowing some customers to more easily stream their content to the AWS Cloud where Veritone will process and enrich their content with AI as it is captured, in near real-time and at scale. For example, a customer that has a lot of content being captured on-premise such as a shopping mall or standalone retail outlet and they want to start running cognitive processes like facial or object recognition for security purposes. This can be an interesting proposition because all of their content is likely being captured on legacy video systems on site and in some cases, the customer isn’t extremely technical,” said Ward.
Commenting on the benefits of KVS, Yousuke Okada, Co-founder and CEO of ABEJA, said that by using the service, his company is able to focus on developing innovative applications and bringing the power of their own AI platform to verticals such as manufacturing and healthcare, where real-time video analytics can generate actionable intelligence. Okada was quoted on AWS’s website saying, “Before Amazon Kinesis Video streams, it was hard for us to perform real-time video analytics, and instead we had to rely on batch driven analytics, which makes insights less impactful. Now, we can focus on building the deep learning solution that power our AI platform instead of worrying about infrastructure for real-time video ingestion or dealing with the variety of camera makers and models.”
According to Okada, ABEJA Platform is a video analytics platform that “mounts deep learning and the whole pipeline for social implementation of deep learning.” He continued, “Deep learning is able to find appropriate features of a specific data from big data without human intervention automatically. Customers in manufacturing, infrastructure, logistics and retail industries have already utilized deep learning technology on our platform. We also provide SaaS for the retail industry to analyze retail’s specific data such as shopper behaviors and optimum shopping experiences.”
However, Okada added, “handling of real-time video data is very complicated and it often takes much time to complete building the applications than we suppose. It will be easier that our customers analyze real-time video stream and many data sources on our platform by using Kinesis Video Streams.”
When asked how the partnership with AWS would change the company’s targeted customer segments, Kattan said that Agent Vi has always focused on developing enterprise-grade solutions for customers who require highly accurate video analytics, such as municipalities and remote guarding companies. “This integration, which will simplify the implementation of our video analytics SaaS to potentially any camera, will allow us to more easily penetrate our current target markets as well as expand to additional markets, including mass market applications which we have not targeted before, like the smart home market,” he said.
Ward said that the partnership with Amazon Kinesis Video Streams and the recognition of machine learning expertise by AWS will have a positive impact on Veritone’s customer reach.
As for ABEJA, building on its success in the retail sector, the company expects itself to branch into other vertical markets. “For example, the customers of the manufacturing industry have already built their applications with deep learning technology on our platform, for failure prediction and automatic inspection, etc. That’s why they will be able to build such applications much more easily by Kinesis Video Streams,” said Okada.
Amazon KVS is currently available in select parts of the U.S.; Ireland; Frankfurt, Germany; and Tokyo, Japan, enabling users to have a video ingestion and storage platform on which to build their vision-enabled applications. With the partnerships between AWS and Agent Vi, Veritone and ABEJA, users can now enjoy more options in video analytics solutions tailored to their needs.