The problem that was never solved for video surveillance, scalability. But now it’s here.
Video Surveillance, Video Management System, Video Surveillance Monitoring, CCTV, and the list goes on. Everything finds back its roots during the second world war. In 1942, the nazis were the first ever, using a video management system to observe the launch of long range guided ballistic missiles.
During, but definitely after the second world war, video surveillance went viral. Hardware manufacturers starting developing advanced video cameras at an exponential speed. This was the start of real innovation, and guided us to advanced topics such as machine learning, which we are already familiar with nowadays.
While new technologies, including hardware and software, were developed over the years. The public and industry started to implement and install video surveillance hardware as well, it had no longer a military function only. It was now available for end-consumers and businesses, with a focus on do it yourself but also exploring better performance, efficiency and new business models for enterprises.
We never solved the underlaying problem, bringing true scale.
Starting with a single video camera or video stream, 80 years ago, we now have thousand of video cameras across different enterprises and millions of video cameras across the globe. Video surveillance hardware scaled rapidly, but unfortunately the video surveillance software didn’t.
It’s interesting to understand and look back how the industry tries to solve the scaling problem, but never tried to solve the actual scale problem. Not sure if you can still follow, but let me explain you.
When enterprise started the implement video surveillance at their different sites, factories, stores, and more, their video landscape exploded rapidly. It was hard to manage, so many cameras, so many sites, so many maintenance, so many streams to be monitored. It is pure chaos.
Guards are sitting in rooms with hundreds of video streams in front of their eyes. At the beginning those streams had to be monitored manually, which means you have to watch a video stream the entire time. The person in front of the streams was the filter, the brain, to decide when something was happening of interest.
This method was and is not scalable, because the brains have limited computing power to processes more than 10 video streams. Once you go beyond that limitation, a human will start making mistakes by not noticing events of interest. To increase this limitation, methods such as “motion detection” were introduced. This increased the range of video streams to be processed. Now you were receiving alerts from a pool of video streams, and no longer had to monitor each individual stream.
Motion detected is a huge improvement but still introduced false-positives. Events such as rain, snow, or other events that made significant changes in a video stream, were also reported and send to a guard. To overcome this the rise of computer vision and machine learning, and more precisely video analytics, brought new perspective in the world of video surveillance. Now we combined motion detection and machine learning to detect specific events such as pedestrians, cars, etc.
To conclude, both motion detection and machine learning scaled video surveillance in the perspective of an end-user, but it never scaled the video management system, it even introduced new scaling issues (such as scaling multiple GPUs). Previously we would hire more people for monitoring video streams, and ended up with some linearity. The more streams you are monitoring the more people you would require. What motion detection and machine learning changed is the linearity of a person versus the number of video streams it can monitor.
What is scale about then?
It should become clear that scaling the linearity of a person versus the video streams is not the same as scaling the underlaying video management system.
While growing a video landscape, we always tried to improve the ability for a person watching more camera streams. However we never scaled the underlaying video landscape.
The real problem becomes visible especially in brown fields, where already a vast amount of video cameras are installed. Adding more cameras to the video landscape, will introduce new complexities such as new NVR’s, new disks, etc. Once entering maintenance mode, new problems will arise: Hard drives goes bad, the entire NVR breaks down, new video cameras might require additional NVR’s etc.
Scale is more then expanding an existing video surveillance landscape seamlessly, it’s also about high-availability, fail-over, auto-recovery, etc.
When a NVR goes down, you want your cameras to be rescheduled on another NVR. If a Hard disk goes down, you want another Hard disk taking covering over, and don’t want to loose any data (erasure coding).
Flexibility the new scale
Next to scale, flexibility became even more important. Machine learning is introduced in many different line of business and processes. Due to this new business cases are being developed, and this opened the door to new video analytics solution and algorithms. For example in retail, calculating the queue time, the number of customers entering a store, or in manufacturing, safety of workers in dangerous environments.
Due to this integration became key, how can you integrate with an existing VMS and develop your own business process? How do you make a customer Machine Learning model available?
Welcome to Kerberos Enterprise
Kerberos Enterprise is a game changer, which tries to solve the real problem: making a video landscape scalable and flexible.
To handle this challenge, Kerberos Enterprise, relies entirely on Kubernetes and comes with a modular and micro service design. This makes true scalability and flexibility possible.
Bottom line it allows you to be cloud agnostic:
- Edge deployments
- Cloud deployments
- Hybrid deployments
Allows you to bring your own storage:
- Edge storage (MinIO, Ceph)
- Cloud storage (S3, Google Storage, Azure Blob Storage, Storj)
Integrate any processes through message brokers:
- Kafka
- SQS
- ..
Due to this and thanks to a Kubernetes based architecture, features such as auto-scale, load-balancing, high-availability are enabled by default. You can reuse your existing computing power, and extend and integrate your video landscape easily with your own business processes.
Are you ready for real scale? At Kerberos.io we are!
