The Device Chronicle interviews Miguel Melnyk on video surveillance as a service (VSaaS) for cost reduction and product innovation in Physical Security.
Northern Tech CTO and Mender co-founder Eystein Måløy Stenberg recently identified three key trends in building management systems . Remote management and AI driving cost savings were two of these key trends. Remote management, AI, cloud, with OTA software updates drive innovation and reduce physical security costs, an essential facet of building management systems.
There is a universal need for increased physical security combined with a desire to reach videos remotely and get valuable customer insights and other business analytics from the videos in the cloud. Consequently, the adoption and growth of Video Surveillance in the cloud has exploded in recent years. Video Surveillance as a Service (VSaaS) enables organizations to access live remotely and recorded video, store, play, forward, manage, and monitor surveillance video footage in the cloud. Videos are not kept on an onsite server, NAS, NVR, or computer hard drive but saved in the cloud. IPTechView projects that the VSaaS market will reach $6 billion by 2026, with a compound annual growth rate (CAGR) of 16% from 2022 through 2026.
Software plays an increasingly important role in video intelligence and physical security: Its influence grows while the hardware gets commoditized, services move to the cloud and OEM and 3rd party developers put firmware and new applications on the AI cameras. OTA software updates are integral to the cloud-to-edge architecture in VSaaS and edge AI cameras.
Miguel begins by explaining that VSaaS is implementing a traditional video management system (VMS) seen in surveillance on the cloud and as a service. The main benefits of VSaaS are those directly inherited from the cloud. Miguel explains that “The cloud application brings elasticity. A customer can use configuration to expand a system without the need to add equipment or wait for equipment to be delivered. VSaaS only requires small amounts of on-prem equipment, so it helps with cost management. There is less capital expenditure (CAPEX) now on on-prem NVRs and VMS servers.”
Cloud also provides flexibility to provide extensive and improved features in surveillance. Since all the cameras feed the video to the cloud, all the cameras in a given setup, multi-location, can be accessed. He says “Consider a bank with 500 branches. An authorized user can see live video from all of them if needed. They can centralize recordings and save metadata and events for those cameras, which helps better manage incident investigations. For instance, when something happens, an authorized user can search all the information in a database consistently and uniformly for those sites.”
Miguel points out that there has yet to be a complete transition from on-premise to VSaaS. On-prem is still viable as the device storage costs have dropped substantially. Storage is available both on camera, in camera in many cases, and on edge devices. The prices have dropped with SSD cards, SD cards on cameras, etc. VSaaS storage costs have also dropped, and bandwidth is much more accessible. A systems integrator that is about to undertake a VSaaS implementation has several options to choose from to ensure they can implement the policies that the company sets, balancing costs and administration efficiency on distributed VSaaS deployments.
Miguel adds that institutional policies for video data handling often dictate the cost of the implementation. Infrastructure availability, the cost of cloud connections to the cloud; and where customers want to store the video are also important factors. Storage could be partly on-premise and partly in the cloud, and several factors determine the cost of managing storage. Some companies must store almost every video they catch for a very long time. That is not possible on-prem. Therefore, in this case VSaaS sends video data to the cloud for longer-term storage. And the last factor is what to save. So as part of the institutional policy, some organizations may say, "Well, I don't want to save any content where movement is not detected at all because it's not useful. So it's a waste of money. In warehouse use cases, if there is movement after 6 pm, the AI camera will detect it instantly and machine-classify it as an essential event. This movement will be recorded and reported.” In summary, there are many options nowadays to accommodate many use cases, and they are all defined and dictated by the policies that the company or institution that stores video data needs to satisfy.
Miguel explains that AI lets VSaaS applications “look inside” the video stream and analyze what's in it and what is happening there. It can detect objects and situations. AI cameras can analyze the video stream they capture on the camera and generate metadata for every event or object they pick up. A metadata stream contains descriptive text in a format compatible with database storage. The system generates a metadata stream alongside the audio and video. It is a third stream that the camera adds to the audio and the video it records. It goes into a database with the recordings and allows authorized users to steer their attention to things. In the old days, when there was no choice but to watch the videos, there were monitors arranged in “video walls” that security guards needed to scan directly to pick up events of interest. With the advent of AI, when the camera detects an event on a specific camera, the focus shifts to that event and that camera and perhaps surrounding cameras in the proximity that may provide information to see what's happening.
The policy dictates what video information should be filtered out. In surveillance, some events always happen. Miguel shares two examples of this: “For instance, if a vehicle rolls into a zone that is not supposed to be in, that may trigger an event. And if there is loitering, that should also direct live video streams and trigger recordings accordingly because that is an event of interest”. So, AI and object detection allow the authorized user to follow what is important. Consider a scenario where a hundred cameras are constantly pumping live video. It's super challenging to watch. And most of the time, events do not happen. So, the authorized user can safely ignore all the captured video. If recorded, they may retain it for some days in case something happens and should there be a need to go back and investigate. AI indicates where to find the relevant events that are security matters. It drives a lot of cost-effectiveness for companies as they develop these policies and only filter out the most pertinent data.
Miguel’s employer Hanwha Vision has created an open platform for video cameras that allows developers to build applications. Miguel explains that open platforms ensure extensibility, especially for complex embedded programming environments. “When you go to embedded environments, it becomes difficult for developers to jump that entry barrier, and open platforms simplify that dramatically.”
Hanwha Vision provides a clean software development kit (SDK) and a set of APIs that developers can easily use to create apps. Developers can expand the functionality of the applications into a specific embedded environment. That has been done successfully by all major embedded device vendors - smartphones, smart TVs, and set-top boxes for cable TV. As a leader in the AI and IP camera field, Hanwha Vision has created Wisenet Open Platform, which has generated an ecosystem of partners that develop applications by using the rich functionality and technology in cameras.
Miguel argues that with the advent of open development platforms, 3rd party software needs frequent updating on the embedded devices. Miguel says, "Before, proprietary platform software was the only thing that could be updated on a camera. Now with 3rd party applications, these must also be updated. For that, an update is required every time there is a new release for an open app. On top of the regular firmware and internal updates required for devices, the device must be updated every time the partners release new versions."
Miguel adds that there is emerging technology that will lead to more frequent software updates. For instance, Custom Object Detection is a novel technology brewing over the last year or so. It will require that companies update that specific application every time they come up with a new object to be detected. Hanwha Vision has developed Flex AI, a machine learning system that allows for streamlined training so customers can train AI to detect objects. AI cameras now go way beyond surveillance, and end customers want to use this advancement to optimize their internal business processes. For instance, end customers can ensure that forklifts in a warehouse do not go outside their allowed operation zone: a detection system can be mounted for a specific forklift model with line-crossing technology and create events if the vehicle crosses the line.
The possibilities seem endless. Miguel concludes, "We've been talking to customers about anomaly detection. Each case may require a different detection model to ensure high accuracy. It would be best if the end customer created specific models for each unique anomaly detection scenario." Another example to illustrate the breadth of use cases is object counting applications. In cattle counting, models are different depending on whether you want to count sheep or cows.
In summary, when a developer makes a new model, the authorized operator must update the cameras in a setup. It is becoming mission-critical. "Ultimately, the software is becoming more powerful than the hardware and more pervasive in many ways, and that will be the future. Absolutely. It is very much software-defined. It follows the same evolutionary path technology has observed in many areas."
Get in touch with Miguel on Linkedin or by email: m.melnyk@hanwha.com.