A programmer's father made his own AI feeding detector to predict when the baby is hungry and prevent his wife's sleep from being affected by his crying.

Caleb, a programmer brother, recently gave birth to a son, but troubles also came with it.

The newborn baby needs to be fed every few hours, and will burst into tears when hungry at night. At this time, the whole family will be woken up, and they will not be able to sleep well after being tortured all night.

He consulted experienced people around him, and the advice he got was:

Just endure it and get through the first few months. .

But as a programmer, Caleb was unwilling to give in and began to use his engineer's thinking to solve problems.

In the end, through a combination of camera AI algorithms, he developed an automatic baby hunger detection system that can detect the baby in time before it actually starts crying.

#Caleb tells the program to send a notification to his phone once it determines that the baby is 100% likely to be hungry.

# In this way, he could get up quietly by himself to breastfeed, and used his technical skills to protect his wife's sleep.

# Such a system does not need to be developed from scratch, because the off-the-shelf human posture detection algorithm is already very mature.

For example, Caleb used Google’s open source MediaPipe, which includes complete full-body posture, facial mesh and hand movement detection.

The remaining question is-

How does the AI ​​determine if the baby is hungry?

Before starting to take action, Caleb first went to the mother and baby forum to read a lot of information.

According to theory, crying means that the baby has entered the later stages of hunger. It is not easy to breastfeed directly at this time, and you need to calm the baby down first.

Early manifestations of hunger also include smacking or licking lips, opening and closing the mouth repeatedly, sucking lips, fingers or other clothing toys, etc.

Caleb wrote codes to give different weights to these behaviors based on his experience in observing his own children.

For example, smacking your lips will have a 10% confidence level, and putting your fist to your mouth will also have a 10% confidence level.

The baby turning his head back and forth indicates that he is looking for food sources. He observed that his baby would turn its head at different frequencies depending on the degree of hunger.

So he set it within a short period of time. The more frequently he turned his head, the more confidence he gained.

During this period, we also encountered situations where a baby holding a pacifier would cause occlusion and the algorithm would not be able to accurately identify the movement of the lips.

To this end, he also retrained his own customized model based on the open source algorithm, giving confidence levels based on the situation of holding a pacifier and not holding a pacifier.

#During this process, he also discovered that the baby would spit out the pacifier when he was very hungry. This action will increase the confidence level by a full 30%, which means you will cry out soon.

After this system was put into use, it has indeed brought many benefits to Caleb’s family. He concluded:

The baby is happier and the adults can sleep more. .

However, this story is not over yet...

Is an automatic feeding system feasible?

The preliminary results of the software did not satisfy this old man’s DIY soul.

Next, he linked this system with hardware and mechanical devices, trying to create a fully automatic breastfeeding system.

The idea is very bold, but judging from the ending, it is still a bit over the top.

He also knew that this thing was prone to accidents, so he first found an adult to play the role of a baby for the experiment.

I saw him holding the pacifier in his mouth and imitating the movements of a baby, turning his head and smacking his lips repeatedly to increase the system's confidence, and finally spit out the pacifier to trigger the system alarm.

What happened next, I can only say "dangerous actions, please do not imitate".

Caleb made a video of this experience and shared it online, which attracted attention in the programmer and DIY circles.

He revealed that it only took about 50 hours to complete the entire development work, because the MediaPipe open source tools are already very complete.

Some netizens commented: If I were to develop this system, my children would already know how to make soy sauce by the time I finish being a child.

There is a reason why Caleb can do it so quickly. He is already familiar with the development process of target detection applications. He has also done dog detection in his own yard before. When to poop.

Some people feel weird about his approach.

When you are hungry → you have to cry → you get food. This should be a necessary stage for learning to interact with the surrounding environment and communicate with your parents. If you can get food just by lying down, will it affect your development?

Another netizen thought that the professional advice he got from parenting courses was to try to feed babies before they cry, so it shouldn't be a big problem.

Caleb himself also said that he and his wife will not completely rely on algorithms to decide what to feed their children, but with the help of AI, they can do a better job raising their children. efficiency.

Of course, there are more interested netizens urging him: open source quickly!

Video address: https://www.youtube.com/watch?v=Lda1Sq8HRY4

MediaPipehttps://google.github.io/mediapipe /​

The above is the detailed content of A programmer's father made his own AI feeding detector to predict when the baby is hungry and prevent his wife's sleep from being affected by his crying.. For more information, please follow other related articles on the PHP Chinese website!