Lego, a building block toy that can fully exercise people’s creativity and imagination, even adults love it. How much can Lego’s heavy obsessives love it? Lego enthusiast Mike Doyle used tens of thousands of Lego bricks to make an abandoned house when the US housing crisis broke out in 2009; New York artist Nathan Sawaya used Lego to create 3D sculptures and large mosaics.
Daniel West, a software engineer in Australia who is also a Lego enthusiast, recently tweeted his Lego creations. Also combined with CNN and 3D recognition, can be said to play to the peak of the point!
It took him two years to produce the world’s first “LEGO Universal Sorter”, an AI-driven automatic sorter that can identify and classify any Lego parts that have been produced.
That is to say, no longer need to worry about generalizing the finishing parts! (Pretend we can own this machine, too)
“Powerful projects and execution! Combine Lego and engineering in a practical way! ”
“Wow Daniel, that’s so impressive! I’m amazed at your creativity, your resourcefulness, and how cool you’re doing. I will introduce your young son to what you have done and use you as a role model for his creativity and perseverance! ”
even successfully attracted the attention of the King.
“Hello, Daniel. I’m evgeny, the owner of one of the largest second-hand LEGO toy stores on Bricklink. I’m interested in this sorter. How can I contact you to discuss it?”
It can be seen that the invention of this little brother is really unusual! So how on earth did he do it?
First-generation LEGO sorters: the inspiration for little brother
Daniel mentions that Akiyuki, a Lego enthusiast from Japan, invented a Lego sorter in 2011, which was his original inspiration. The sorter at that time, though not so gorgeous, was also very powerful.
Akiyuki had fewer database images at the time and was slow to recognize and sort.
Today, Daniel’s machine consists of more than 10,000 LEGO parts with six LEGO motors and nine servo motors to power conveyor belts and agitators.
The ability to sort nearly 3,000 LEGO parts into 18 different boxes, and sort one part every 2 seconds, which is efficient!
No wonder Daniel calls his sorter the world’s “first LEGO universal sorter” because it uses state-of-the-art artificial intelligence technology to identify and classify any Lego parts that have been produced, even parts never seen by a machine, through convolutional neural networks.
Daniel also plans to publish the code in the future.
What it works?
For the past two years, Daniel has been designing and manufacturing machines that can identify and classify LEGO parts, using more than 10,000 LEGO parts. With computer vision algorithms, any LEGO parts can be identified.
At the heart of the sorter is “Capture Unit”, a small space with conveyor belts, cameras and lights. It doesn’t seem like a big deal, but there’s a lot to pay attention to if you want it to get it done.
The camera takes Lego parts along the conveyor belt and uploads the photos to a server running the AI algorithm to identify the parts from thousands of possible LEGO elements.
The core problem is the need to convert the real-time video stream of the conveyor belt into a separate image of the various parts that the neural network can recognize.
The final goal: from the original video (left) to the evenly sized image (right) and then sent to the neural network. (Motion picture speed is about 50% slower than real-time video)
Target detection is used here to detect the existence, location, and size of the target so that the part can generate a bounding box at each frame. On the face of it, it looks simple, but it’s actually difficult.
In order to achieve the smooth identification and sorting of parts, there are many areas to pay attention to, such as the position and angle of the camera, the light source to ensure adequate, and parts can not be the same as the color of the conveyor belt, otherwise can not deduct the background, the belt color chosen by brother is pale pink.
It also takes a lot of time and effort to train neural networks. Think of it as a virtual brain that can accomplish a specific task by accepting input and converting it into the corresponding output.
In general, the more data you enter into a neural network, the better it is able to complete tasks. The neural network of the LEGO sorter outputs the number of the LEGO part by entering an image of the LEGO part.
Because of the hundreds of types of LEGO components, they come in a variety of colors and are shaped differently from different angles. Therefore, collecting the right training data set is the hardest part of the job. Brother captured 300,000 images after the sorter ran a few days later, part of which.
One thing all Lego lovers have a headache
That’s right, it’s sorting! Daniel has this inductive finishing hand, no longer have to worry about that bucket of parts! But we also only have an eye-opening part ah, thinking of those years were piled into the bucket of Lego, or had to sigh a long sentence, alas, on how many kinds of parts Lego so far, really can not find the exact number, because the classification is not the same, the kind of statistics are not the same, and it seems that every day a new Lego parts are born… Lego enthusiasts have to communicate with each other, explore the experience of finishing Lego parts, digestbacteria looked at the pods and know the vast number of netizens enthusiastic sharing, the general classification method has by size, by color, by series, and even according to the number of particles on the parts. Parts are filled with utensils are more than some storage boxes, finishing boxes. Ah, I’d love to hand the job to AI. Some netizens joked in the comments section of Daniel’s video that they’ve included the production of the AI Lego sorter in his retirement plan.
“It’s really great, are you going to rent it?” I have thousands of parts in my attic, and it’s a nightmare. This is included in my retirement plan! “