Neural Labs - A trip to the outer edge!

By Elías Valcárcel, CEO & Co-founder

THE WEIGHT OF HISTORY

Historically, Neural Labs has always focused on processing with centrallized servers.

Our license plate recognition systems were installed on servers and received video streams from conventional cameras; we performed the processing, image analysis and neuronal analysis to classify the characters.

Centralization has provided us with many benefits in an infinite number of scenarios, for example:

- Easy maintenance, with a single point of configuration (fewer points of failure).

- Possibility to use cameras from any manufacturer (even combining them within a projecr).

- Use of very simple, affordable cameras.

This approach also had, currently has and will continue to have drawbacks, such as:

- High bandwidth usage (although to be honest, there has usually been more than enough bandwidthe available).

- The need to use large - and sometimes expensive - servers.

- Space needed in the data center.

Neural Labs has installed this centralized architecture in countless cities, providing access control and more around the world.

PARADIGM SHIFT

Around three years ago, we detected that the market was changing, moving towards a more distributed approach.

- More and more bidding specifications required edge processing, using "smart cameras".

- The simple cameras that we had used in centralized processing already had enough processing capacity to run a high-performance license plate recognition algorithm (with no price increase).

- Several camera factories in China were beginning to offer embedded license plate recognition products at an acceptable price and quality.

Within the company, we started to consider migration towards the edge, where we were met with several challanges.

CHALLANGES

A new architecture

One of the most important milestones was transporting our license plate recognition algorithm from an Intel architecture to an ARM architecture, the processor on which many of the SoC (system on chip) systems are based.

One of the aspects that sets Neural Labs apart from the rest has always been the proprietary nature of our video analytics algorithms. All have been developed according to the C++ standard and by our own R&D department.

This made it much easier for us to transport and optimize our engine to this new platform.

The solution

The second challenge was to develop an applicarion that would run inside the camera and contribute business logic to license plate recognition.

Clients do not want a simple license plate reader. Clients need a barrier that is activated by this reading, according to a list of authorized vehicles, for example. Solutions!

We successfully overcame this second challenge, where the necessary technologies and the limitation of the processor required experience that we did not have, but very quickly acquired.

Homogeneity

All products on the edge had to be seamlessly integrated into Neural Labs' ecosystem of existing solutions. There could even be HYBRID centralized/distributed solutions that were transparent to the client.

The sharing of protocols (all our products use the same ones) and the inclusion in our centralized system of an EXTERNAL CAMERA that returns the data already processed (ideally, our Ghost and EDGE models) permit:

- The interconnection of our solutions, which can handle 100% centralized, 100% distributed or hybrid projects.

- Our clients only need to be integrated once into our protocols, and then they can use any of our products.

- Any data recognized by Neural Labs is accessible from our WEB platform, regardless of its origin. Making the origin (product) 100% transparent to the client.

Today, Neural Labs has a solution for any scenario, regardless of the architecture required.

PRODUCTS

The first product on the edge was paradoxically the most difficult one, as it is intended for very high-speed scenarios. Neural GHOST is a product capable of recognizing license plates at up to 180Km/h. This is possible thanks to the acquisition characteristics (camera, lens, lighting) and the processing capacity.

This product has been very well received and is now being used in ITS, traffic violations, low emossion zone management, etc.

For the second product, we took the plunge into a radically opposite scenario: access control.

For this scenario, we developed Neural Edge, a license plate recognition engine, accompanied by an entire access control management application, based on lists and actions.

This product works with normal security cameras, as long as they meet these requirements in terms of acquisition and processing capacity.

Once again, this more recent product is being very well received in 2 areas:

- Companies with vertical products that only require our data (the license plate).

- Companies that want to use the full logical capacity of the product to automate access control.

The third product is one that bridges the gap between the two: URBAN Edge, a product that covers the city market segment.

This product is capable of recognizing license plates at up to 90km/h and also works with standard security cameras. Urban Edge is strongly focused on security and mobility applications.

THE FUTURE

With the boom of smart cameras, Neural Labs has been able to adapt to the new reality by enriching our offering, adding the possibility of edge processing, when necessary.

Centralized and distributed systems are sure to coexist for many years to come, and the trend may even go back and forth, as it did with computing. In any case, the company is prepared to offer the best solution for each case.

Neural Labs will continue as always, improving each and every one of the products (iterating), improving interoperability with other systems, in order to remain faithful to the company's DNA of always providing highly reliable, robust and open systems.

Elías Valcárcel, CEO & Co-founder