The use of artificial intelligence in conjunction with driver assistance systems is now becoming increasingly common. Another exciting use for AI is in "People Sensing”: when people are entering shops or buses/trains, a camera can use image processing to detect them entering. The advantage? Businesses, for instance, can keep track of how many people are in their shop and can stop people entering if required. This is a particularly indispensable technology, especially in the age of Corona. The Berlin company HELLA Aglaia manufactures high-precision people-counting devices in this field. Its portfolio also includes driver assistance systems and other exciting AI solutions. We spoke to its Managing Director – and native Berliner – Kay Talmi.
Dear Mr Talmi, HELLA Aglaia, as a subsidiary of the automotive supplier HELLA, offers a wide range of technical and intelligent solutions. How did this spin-off company come about? And what’s your personal background?
In fact, it wasn’t a spin-off, but a purchase by HELLA. Aglaia was founded in 1989 with the mission to make people's everyday lives easier and safer by providing visual sensor solutions that, simply put, enable machines to see. A milestone was the series production of "Blind Spot Detection" for monitoring the blind spot. By 2006, we’d grown to about thirty employees.
After the takeover by HELLA in 2006, the development of a front camera for vehicles led to the creation of further safety-relevant functions such as traffic sign recognition, high beam and lane keeping assistance. From the very beginning, HELLA has supported us intensively and has worked with us to expand the company and its portfolio. In the meantime, we have grown to 450 employees. Looking back, working with HELLA is a very beautiful symbiosis, which makes me proud and grateful.
I studied computer science myself at the TU Berlin. Afterwards, I started my own business with a company specialising in image processing. Besides my work as managing director I was responsible for the planning and implementation of image processing projects in the field of human-machine interaction: a camera was used to operate a PC, on the one hand for controlling games, but also as a method of interaction for the severely handicapped. The idea was exciting, but in 1997 it was unfortunately too early for this topic. Then I worked for a fingerprint recognition company in Berlin, among others.
I joined Aglaia in 2004 and was initially responsible for the production launch of the Blind Spot Detection System (BSD) and soon for the entire development. Since December 2009 I have been managing director of the company. Thanks to my background I am still very deeply involved in our topics.
One of your solutions is called “People Sensing” and can be used, for instance, for counting customers in shops – a practical solution, especially during Corona times. How exactly does the artificial intelligence behind it work?
People Sensing is the name of our business unit, which has been involved in automated people counting in retail and mass transit for over 10 years. Counting people does not only work with artificial intelligence, but primarily with image processing. A video camera is fitted with two cameras that function like two eyes. It’s mounted on the ceiling and estimates the 3D information between them via the two cameras. Here, you can imagine a crowd of people as a forest where people’s heads are seen like treetops. The camera can very precisely recognise and follow these treetops within an area of several metres, approx. 6x6 m, via the 3D information. By linking several cameras, this area can even be extended to more than 500 square meters. AI is then used to make the whole thing secure and for very precise tracking. Thanks to the use of AI, we can achieve high standards in people recognition.
Often we have the obsticle that other objects are in the picture, for instance when someone carries a high backpack or an umbrella while shopping, which is also recognised as a treetop initially. At this point, the AI can then very precisely recognise what is actually visible on the camera, thus achieving high quality results.
We do all this not only in the retail sector or public buildings, but also for instance in the bus and rail sector, where there is a very high demand for counting quality of up to 99 percent. For example, if a school class enters a bus and many of the pupils are carrying backpacks or umbrellas, then People Sensing must meet this 99 percent level on average. We achieve this primarily by the use of smart AI algorithms.
In what area is your People Sensing primarily used?
We have a very strong presence in the bus and rail sector for over ten years now, where we’re also one of the market leaders worldwide. In the retail sector, the topic is currently gaining momentum, thanks also to Corona. Currently, there is also a requirement in shops that only one person per 10 square metres is allowed in a shop, which can basically only be measured properly using a counting system. Therefore, we are currently receiving many inquiries. For example, stores need an access control system or a so-called "traffic light function", which jumps to red when the limit of people in the store is reached. Here we work together with a big network of partners who take over our People counting sensors and then, for instance, build such a traffic light system from it and integrate it into retail chains afterwards.
To what extent does your solution replace staff and in which areas can AI and humans work together even better using it?
The pure counting function at the door can really be completely replaced by our sensors. Especially if a lot of people walk into a shop or a bus, for example, it becomes almost impossible for one person to count them excatly. A machine, on the other hand, looks at everything from above and can count neutrally around the clock and in real time.
In a lot of areas, however, where our “counter” is used, there is an interaction between the person who evaluates and assesses the results and draws the right conclusions, and the machine. AI and humans therefore complement one another optimally.
How is the data recorded by “People Sensing” handled? And how is data privacy guaranteed?
A big advantage of our system is that we do not record any data, i.e. no images. This anxiety often exists, and we talk to our customers a lot about it, but we’re able to completely dispel their worries. Because of the 3D information that our sensors record, we only have the height and the directions of the movement as an information, but no photo of them. In addition, each piece of information is only evaluated once, i.e. counted, and then thrown away. Personal privacy is thus maximally secured.
In addition to your people counter sensors, you also sell, among other things, driver assistance systems, which have now been sold to VW. How does this software work and what advantages does it offer?
We’re using AI a lot with this system i.e. neural networks that recognise different objects. Among others, we currently have the so-called trace recognition in our portfolio: for example, if you take your hands off the steering wheel while driving a car, your car will nevertheless stay in the middle of your lane. This is now very common.
Our second topic is traffic sign recognition: here, AI recognises the current speed limit in force. Navigation systems do recognise these to a certain extent, but when road works signs appear, for instance, the front camera recognises this much sooner than the navigation system. In combination with the navigation system, it’s then possible to ensure that the prevailing speed limit is observed.
The third function we implement is pedestrian detection. This function is very important and has a high safety standard. The car brakes at the shape of a pedestrian, sometimes even to the point of emergency braking. With the help of AI and neural networks, the pedestrians are then recognised.
We have the challenge with all these functions that our world is simply very complex, especially the inner-city environment. A pedestrian can e.g. appear in many different ways – for example be wearing a hat, pushing a stroller or moving in a funny way. This complexity was very difficult to face with previous methods, because you had to program a rule for every different pedestrian. However, using neural networks this can be solved: you show the machine as many examples as you like and it then learns the rules itself. This has been a huge advance and makes it possible for us to implement many functions. AI is indispensable to coordinate on the high variability of objects and to guarantee high quality.
We develop the software for this purpose at our location in Berlin, which we sell to our customers, who in turn build their camera around it. VW will then use and further develop this in the future.
But we will continue to develop AI here in Berlin. On the one hand through the People Sensing team, but also with other topics - outside of cameras.
What other areas are you currently planning in? What plans does HELLA Aglaia have for the future?
We are currently working on the development of intelligent and safe control systems for high- and dual-volt batteries, in short: on the exciting topic of electromobility. We will also expand People Sensing and look for other markets for our product solutions too. There are some very exciting examples of this, such as building automation or building management, which can also be supported by a people counter. For example, if I know how many people are in a meeting room, I can control the heating or other things in the building.
We will act as an internal service provider for the HELLA Group worldwide, but will also market our AI solutions externally to third parties in the future. A typical application for AI is quality monitoring and improvement in production. Here we develop application-specific solutions to increase the cost efficiency, profitability and quality of products and thus of the company.
We will also be active in other domains, for example in controlling data with the aid of neural networks or when checking legal texts. There are many possible applications for artificial intelligence, which we’ll monitor with a team at HELLA Aglaia.
What advantages does your Berlin location have over others in the field of artificial intelligence?
There are various aspects to this, one of which is recruiting: it’s much easier for us in Berlin to get good people in the AI field than in other places. On the one hand, we’ve got lots of universities where people come from, and on the other the city is also attractive and exciting for newcomers from abroad. And third, we have the big start-up environment here in Berlin. Especially in the field of AI, there is an incredible amount happening here, which benefits us enormously.
A year and a half ago, we e.g. founded Drivery, the first subsidiary of HELLA Aglaia. There we’ve created an ecosystem around mobility for start-ups, but also for medium-sized companies, suppliers, investors and politicians. This is a place where people can create innovations together on an equal footing. We’ve benefited from the start-up environment in Berlin here too: around 70 percent of the start-ups in Drivery work on AI topics.
Just how prominent this topic is can be seen, among other things, from the fact that the 12,000 square meter area in the Drivery has developed very well in terms of capacity utilization despite Corona.
Especially in the crisis, many companies have become interested in cooperating, also owing to the complexity of the issues. In my opinion, cooperation should also become much more important in German industry in general. Personal exchange and a pooling of energies is still happening far too little. But Corona has been very supportive of people’s openness.
How do you think AI will develop in the coming years? Are we already much further than we thought, or are there still lots of innovations ahead of us?
Indeed, a lot has already happened in the last few years. Step by step, the topic of AI permeates almost all areas of our lives. However, one should not overestimate the current state of technology.
However, we are still a long way from having a similarity to human intelligence. There are certainly areas where AI works well and supports us humans or makes our lives more comfortable and safer. Nonetheless, we’ll probably have to keep working over the next few decades to ensure that it really does become what we imagine artificial intelligence can be. I believe there is a great deal of potential, but we also must invest massively. There are more and more issues that can be solved with AI. Therefore, the technology has a lot of potential. Nevertheless, we must approach the issue responsibly. Germany must also invest even more in research and development in order to remain internationally competitive and to be able to take a steering role. We should play an even more active role in shaping the future, shed our fears and use technology profitably.