A robot will be developed that finds and points out the 'Wally's Find!' Wally in as short as 4.45 seconds

A robot specialized only for finding a Wally from " Wally's Writing! " Was developed. This robot says that you can find the Wally in 4.45 seconds in the shortest time after checking the page.

This robot uses AI to find Waldo, hence ruining Where's Waldo - The Verge
https://www.theverge.com/circuitbreaker/2018/8/8/17665268/wheres-waldo-finding-robot-google-cloud-automl-ai

The way the robot actually finds the Wally from "Wally's Writing!" Can be confirmed with the following movie.

There's Waldo is a robot that finds Waldo - YouTube


This robot performs image recognition using AI, and Google's Cloud AutoML Vision is used for AI learning. First, upload various images of Wally collected in Google image search to Cloud AutoML Vision ......

Let AI learn "What is a Wally?" For AI's learning, 62 facial photographs of the wally and 45 pictures showing the head and body are used.

The robot is equipped with a camera for face recognition, and a metal arm with a hand to point to the Wally. It seems that these are controlled using Raspberry Pi .

I actually move the robot. Set "Warley!" And move the robot ... ...

The camera that was attached to the arm for the first time photographs the picture of the page actually opened.

After that, the robot extracts only the face part of the person from the photographed photograph, and searches for the person corresponding to "Wally".

Then, I found a person who seems to be a Wally with a probability of 97.5%.

Immediately after that, the robot arm pointed to Wally as much as to say "This is the culprit.

Even if I open another page ......

I found the Wally soon. It seems that it will not operate only on a specific page of "Wally search!"

This robot was developed by Matt Reed who works for adproduction company redpepper . Mr. Reed says, "I thought that the number of images used for machine learning was too small, but I was surprised with high precision as I thought." Mr. Reed did not have experience developing robot using Cloud AutoML Vision, but he said that it could be realized in a short period of about one week.