At the "AI Edge Conference & Solution Contest 2021" held by Oki Electric Industry Co., Ltd. (OKI) in December 2021, "Human Gaze Visualization", which can easily simulate human gaze A wide variety of ideas were selected, such as "sight simulation AI" that detects "urine" and "feces", and "preventive maintenance and monitoring by visualizing" odors "which is useful in the medical and nursing fields.
This contest uses OKI's AI edge computer "AE2100" to compete for ideas and technologies to solve social issues in various industries. Prize money is set for the finals, with 2 million yen for 1st place, 1 million yen for 2nd place, and 500,000 yen for 3rd place. There are also supplementary prizes such as media exposure.
Why was such a wide variety of ideas born? What kind of prospects do you have in mind for the future? In this article, we interviewed the people in charge of the companies that won 1st, 2nd and 3rd place in this contest.
Related article: AI contest with a total prize of 3.5 million yen AI that visualizes human line of sight wins 2nd place AI that detects urine and feces I have to work on what I haven't done yet."The day of the contest. Mr. Yuta Machi, President and CEO of Rabbit Co., Ltd., and Ms. Yurina Murai of Rabbit Co., Ltd. (from right to left)
Among many companies, the one that won first place was Usagi Co., Ltd.'s 'Gaze Simulation AI' that visualizes the human gaze. AI that learns a large amount of gaze data collected from eye tracking devices can easily simulate human gaze simply by uploading images and videos to the system.
This solution passed the top in the contest qualifying rounds, and in the final round, "I thought, 'Is there such a way to use it?' It's also very practical." ) were highly valued. Why was such a unique solution born?
──First of all, please briefly introduce your company.
Mr. Murai: We are a company that has been focusing on system development and AI development since about 10 years ago, when AI was not popular. This year marks the 15th anniversary of the founding of the company, which was established by President Machi while he was in college.
Based on our wealth of knowledge and achievements, we support various companies in solving their problems. Past achievements include building an AI system for the Patent Office and providing an image recognition system for NHK's live broadcasts.
Demonstration during the contest. This solution uses a heat map and a green ball to show where people's eyes go when they watch an image. In the heatmap, the red part is the part that people are likely to see, and the green ball shows the part that attracts the most attention in the heatmap. When the employee was actually reflected, a green ball was positioned on the employee's face. Click here for the video
──This time, your company was recognized for its line-of-sight simulation AI and won the championship. What are the differences, or strengths, between conventional eye tracking devices and your company's solutions?
Mr. Murai: The line-of-sight simulation AI uses an AI model that has learned device data to estimate a person's line of sight. I think it's strength is that you can perform simulations many times without worrying about cost or time, because you can drastically cut costs such as device costs, labor costs, and long research periods that were previously required.
──Currently, I don't think there are many solutions that focus on line of sight. I'm curious as to why you decided to take a unique approach that is a little different from other companies.
Mr. Machi: Currently, various companies, including GAFA, are working on AI technology. Considering that the number of functions provided at low cost in the cloud is increasing rapidly, I always think that a small company like ours must work on things that big companies don't do, and things that are unusual. increase.
While there are many companies that use cameras for vision among the five senses of humans, we are thinking about how to show the value of our AI while incorporating elements other than video. I was.
──I heard that development and research began around 2018, and although it took shape in the first two months, you continued to work on improvements after that. Do you have any prospects for the future?
Mr. Machi: This is a common problem with machine learning in general, but I would like to reduce the cost as much as possible, increase the speed, and improve the accuracy. I would like to increase the number of functions that are easy for users to use, and I would like to make it a tool that will be loved and used a lot.
In addition, I would like to simulate the line of sight of people with low vision or color blindness, and aim to become a support tool for creating designs that make life easier for many people.
AI that detects "urine" and "feces" by smell "matches about 90% with nursing care records, isn't the judgment method wrong?"
Contest day . Mr. Tsuboi of OKI, Tokai Electronics Co., Ltd. Marketing Headquarters Sensor & Electronics Emerging Department Manager Koji Hara, Tokai Electronics Co., Ltd. Marketing Headquarters Sensor & Electronics Emerging Department Sakura Kinose (from left to right)
Second place is "preventive maintenance and monitoring by visualizing odors" by Tokai Electronics Co., Ltd. In the medical and nursing care fields, AI can determine the duration and humidity of diaper odors, and detect and distinguish between urine and feces. By analyzing the rhythm of excretion with AI, we aim to escort to the toilet before excretion. Why did the company focus on "smell"?
──First of all, please briefly introduce your company.
Mr. Kinose: Our head office is in Sakae, Nagoya. Founded in 1945. Semiconductor devices account for approximately 60% of the products we handle (*consolidated sales composition ratio by product), and we also handle electronic devices, high-performance materials, and systems.
By market (*sales ratio by market field), automobile-related sales account for about 70% of our sales, given the locality of Nagoya, and we also have many customers in FA/machine tools and information and communications. The sales ratio of the medical field, including the odor sensor for diapers that we developed this time, is still not high, but we would like to do our best in the future.
──Where did the idea for the diaper odor sensor you developed come from?
Mr. Hara: At the hospital, I heard a voice from the facility side saying, "There are so many falls accidents that occur when patients go to the toilet." Patients feel uncomfortable with diapers after excretion, and they stand up and fall or get injured. In the first place, we thought that if the diaper could be changed immediately after excretion, the patient would not have to stand up and would not fall or get hurt.
──There may not be many ways to detect changes in diapers, but I'm curious about how you focused on "smell".
Mr. Hara: Initially, we started by using a capacitive sensor that reacts when the diaper gets wet. At the time, I thought it was a good idea, but I ran into issues such as "Isn't it uncomfortable for the person being cared for to wear the sensor?" and "Should the sensor be replaced every time it gets dirty?"
When I started thinking about odor detection, I was introduced to Koa Co., Ltd.'s odor sensor. I thought, ``If this is the case, you can just put the sensor on the outside of the diaper instead of putting it in the diaper,'' and we started talking about introducing an odor sensor.
Demonstration during the contest. When I sprayed the sensor with air containing a deodorant sheet, the monitor immediately displayed "abnormal". In addition to the odor level, the sensors also detect CO2, temperature, humidity and atmospheric pressure. Click here for the video
──How does the odor sensor distinguish between urine and feces?
Mr. Kinose: AI analysis is used to distinguish between "urine" and "feces" based on the duration and humidity of the smell. In the case of "urine", the smell does not last very long because it is quickly absorbed by diapers. On the other hand, in the case of "stool", it remains in the diaper, so the smell lasts. The evaluation results at the facility match the actual care records by about 90%, so we believe that this judgment method is not wrong.
──Do you have any goals for the future?
Mr. Hara: Our goal is not only to solve the current problems such as "communication is cut off" and "sensors are dropped in the toilet and submerged". It is to hear the voices of hospital staff and other caregivers saying, "This has made my job easier," and from the people receiving care, "I can live with peace of mind without feeling uncomfortable." I would like to proceed with the on-site evaluation firmly so that such voices are born.
Simultaneous inspection of up to 12 meters with a camera "Aimed to help small and medium-sized enterprises take the 'first step of digitalization'"
Contest day
The 3rd place is "AI anomaly prediction by collective data conversion of multiple analog meters" by Metro Co., Ltd. The analog meter attached to the control panel (*) can be photographed with a camera, and the value of the meter can be read/visualized. It can be installed without stopping equipment, and up to 12 meters can be inspected simultaneously. Why did the company focus on control panels?
(*) A device containing various electrical devices for electrical control of machinery and equipment.
Mr. Hideyuki Ishida, General Manager, Business Solution Department, Metro Co., Ltd.
──First of all, please briefly introduce your company.
Mr. Ishida: Our company celebrated its 50th anniversary in 2021 last year. We offer a wide range of ICT services and solutions, including cutting-edge sensing, in-vehicle model-based development, ETL and other data utilization services, high-speed services for supercomputers, and development services for compilers. For details, please see the "Handling products / services" page on the official website.
Mr. Makoto Sugimoto, Manager, Social System Group, Business Solution Department, Metro Co., Ltd.
──This time, your company developed a control panel solution and won 3rd place in the contest. In general, I think that the word control panel is not very familiar. Why did your company focus on control panels?
Mr. Sugimoto: We have heard from our customers that "control panels are installed in various places, such as in the sea and in the mountains, and people check them visually."
In the process of "going on a business trip to the site, visually inspecting, and diagnosing any abnormalities", the person in charge is wasted on travel expenses and personnel expenses, and due to oversights and delays in discovering failures, etc. It may lead to further damage to society. I wanted to come up with a solution that would help me overcome these challenges.
──There are many control panels that were installed several decades ago and have not been recovered or improved. I think that the point where this solution was evaluated highly is that it is possible to digitalize the meter by photographing it with a camera, where people had to visually check it in the past.
Mr. Ishida: Probably, not only large companies but also small and medium-sized factories have meters such as LED lamps and flow meters. Our aim is to provide solutions that are extremely inexpensive and have low installation hurdles while allowing small and medium-sized enterprises to make use of their existing equipment, so that many companies can take the first step toward efficiency or digitalization. That's it. Without such efforts, I feel that Japan's digital rate will not rise easily.
Demonstration during the contest. We appealed how normal data is used as learning data to expose abnormal data. Click here for the video
──On a more technical note, one of the strengths of this solution is the ability to check up to 12 meters simultaneously. How did this system come about?
Mr. Sugimoto: In the contest demo, you can see how the AE2100 is used to capture multiple meters with a single camera, read and visualize the values in the meters, and determine abnormal values and signs. introduced. During the simultaneous inspection, each meter part is cut out from the video, and the position of each needle is analyzed and digitized in real time.
──Lastly, please tell us about your future prospects.
Mr. Sugimoto: First of all, we will brush up the AI inference library (a meter value inference algorithm based on image analysis) developed in this contest to improve its accuracy. After that, I would like to bring the equipment to the customer's site, analyze it, and incorporate the points that need improvement and requests. I have already received several business negotiations, so I would like to proceed with that discussion.
Related article: AI contest with total prize of 3.5 million yen AI that visualizes human line of sight wins AI that detects urine and feces second place