What’s Ahead for AI: Updates from the Hatcheries
Through the Butler Launch Pad, the Blank Center awards graduate and undergraduate student entrepreneurs access to professional and semiprivate workspace to grow their businesses. Known as the hatcheries, these spaces encourage ideation and collaboration. In a series of blog posts, this semester’s hatchery teams will go beyond the four walls of their offices and share their experiences, advice for other entrepreneurs, and industry expertise with us.
The following post is written by Takashi Kashimura MBA’19, founder of 7me systems
Potential of Image Recognition
My venture named 7me systems is going to offer a technology using Image Recognition AI to enhance the learning experience. I am an engineer with over 8 years of experience with Honda R&D in Japan. Now, I am focusing on image recognition AI, which is very different from my expertise of control systems of the automobile, because of the following reasons.
For a control system, a sensor matters. And the system cannot properly work without accurate data from the sensors. Because image recognition technology can sense what traditional sensors could not sense before, it would open a lot of doors for control system engineering. For example, by using a traditional sensor like an infrared sensor, the system can understand that there is something in front of the sensor. If the system also uses a radiation thermometer, it can recognize that the something is a living creature, although it cannot tell whether the object is a dog or a cat. On the other hand, image recognition can identify the object as a dog, not a cat. Moreover, it can even tell that the dog is a Labrador retriever. The difference between the traditional sensor and image recognition is huge. And because the cost of the technology has significantly dropped recently, we can apply this “eye” to whatever we want.
Prior to starting my venture, I created an original image recognition AI, which runs on Raspberry Pi, a cheap credit-card-sized computer. If the AI system detects a moving object in a video image and recognizes it as a person, it will notify me via email and save the image to the cloud storage. The machine learning took 30 hours in total and was done with Amazon Web Service, using 40,000 pictures that I collected. Now, the AI is able to identify dogs and cats as well.
The photo above shows that my AI recognized me as a person with 99.71% probability. It cost only 150 dollars in total, even though all hardware including peripheral devices and the environment of software development had to be prepared from scratch. This shows how much the cost of technology has decreased.
There will be many applications with cameras connected to the internet and cloud computing to process images. By combining the 5G network with 1-millisecond end-to-end round-trip delay, even autonomous driving vehicles can be controlled using this system structure in the future. You will see the huge potential of image recognition as a sensor.
Now, I am working on new software with image recognition AI to increase the learning experience of students and help faculties. I have already developed the prototype myself and proved it works as expected. I hope my product will be widely used in the future and be able to contribute to people’s lives.