[Editor's Note] Artificial intelligence (AI) technology is becoming increasingly sophisticated and is on the rise in terms of commercialization. Autonomous driving has received considerable attention and is currently at the center of the trend, but various services such as facial recognition, object detection, and AI chatbots are becoming more sophisticated and are making their presence felt in the market. We met with AMC PLANET CEO Kim Hyeon-seok, an AI expert, to ask about fire detection solutions using computer vision and hear about the related market outlook and AI development.

CEO Kim Hyun-seok received his Ph.D. in electronic engineering from Kyungpook National University and is conducting research in the fields of network protocols, deep learning, and artificial intelligence. Based on his experience working at Samsung Electronics for over 20 years, he is currently serving as CEO of AMC PLANET, which develops artificial intelligence solutions, and is advancing facial recognition solutions and fire detection systems using computer vision.

■ What made you interested in artificial intelligence?

I got a degree in AI. My previous job was in the mobile division of Samsung Electronics, but since I was only developing software there, I reached a certain limit. When software seemed to have run out of things to do, it met artificial intelligence.

After I got into this field, I had more to do. I thought, “There’s more to do!” and continued to study artificial intelligence. The first thing I did was watch videos from Stanford University and was inspired. I thought, “Ah, this shouldn’t stop at academics.” So I moved to the Bixby development team while I was working at Samsung Electronics.

While working there, I realized that I wanted to create the kind of artificial intelligence I wanted. So in 2020, I came out, studied more, and started my own company.

■ What are the advantages of fire detection solutions using computer vision compared to sensor-based solutions?

If we look at the ceiling, there is a sprinkler and a heat sensor. The heat sensor detects that heat has to rise to a certain temperature to detect it, and then the sprinkler goes off.

This means that the fire has already spread. If you contact the control center in that state, it will likely be too late. It will be too late for the fire trucks to arrive and try to put out the fire.

There were many stories like that, so we started developing this when we received an inquiry from KPOL, a Korean camera security company, and we decided to collaborate because we had researched it.

Because sprinklers have many errors and sensors have many errors. In addition, since most of them have passed their quality warranty period and are malfunctioning, it is said that many cases of fires are occurring.

Sensor-based fire detection systems are expensive to build, but they are also difficult to maintain. I understand that the warranty period for sensors is only 1 to 1.5 years. And they need to be replaced about once every 3 years, so the cost continues to increase. However, the sensing range is not that wide.

It is difficult to maintain it while continuously replacing it. If this is regulated, the company will try to install it within the scope of not violating the fire code, but in reality, it is not easy for the company to maintain it and continue to pay money.

In the case of our fire detection solution, it can be used as is with existing CCTV. You only need to upgrade the DVR (Digital Video Recorder). Detection is possible immediately by simply adding an algorithm under the existing control system.

In the domestic market, CEOs and businessmen have a good understanding of fire detection. So they use a lot of various sensing devices and the fire department also does a lot of detection, so our country has relatively few fire accidents, but in Vietnam, Thailand, and Indonesia, the situation is very poor.

So, if there is a fire in that country, the business owners will completely fail. We have had a couple of discussions with companies in Indonesia about this, and they are very positive about it.

■ What AI algorithm technology was applied to the fire detection solution developed by AMC Planet?

The fire detection solution was created based on a general CNN (Convolutional Neural Network). Of course, a neural network is not limited to just CNN, so everyone knows that the solution currently most widely used in object detection is YOLO (You Only Look Once).

However, the advantage of YOLO is that it detects objects very quickly, but the recognition rate is a bit low. An improvement on this is called GhostNet. When GhostNet passes through an artificial intelligence neural network with different filter types or tendencies, it is classified according to tendencies. The classified data is not passed on to the next layer as is, but is reclassified. It is classified by grouping the same tendencies and passed on to the next layer in a classified state.

Usually, when passing through a layer, a specific feature comes out and that feature is sent as is to the next layer, but in the case of GhostNet, the feature is classified again and a map is created again. Since it is then sent to the next layer, the feature can be distinguished more clearly.

So the recognition rate also increases a bit. Even if you reduce the number of layers a bit, the performance increases even more. There are quite a few papers that combined YOLO and GhostNet rather than YOLO.

If we just apply that, it would be difficult to achieve more than 99% in a solution like ours, so we slightly modified the convolutional layer itself and added the commonly used Batch norm (Batch normalization). Then, we apply a different loss function to each feature and train it through Back-propagation.

The results are currently showing a detection rate of over 99%. This is what was announced in the paper (Volume 29, Number 3, Journal of the Information Processing Society, September 2022).

Continued in Part 2

Meanwhile, on February 2nd at 10:30 AM , e4ds' own webinar AI Tech Concert will be held with Kim Hyun-seok, CEO of AMC Planet, on the topic of AI fire detection system using computer vision .