Friends of basketball should know that walking is one of the most common violations in the game.In order to better monitor whether players walk in basketball games, a little brother named @Ayush Pai(Let’s call him Brother AP)Created an AI referee. As you can see, the computer “watches” the basketball player all the time and can immediately tell if the person is walking.
This AI basketball referee quickly attracted a group of netizens to come and watch.
Some people joked that if the NBA used this AI referee, they would be finished.(Because NBA referees sometimes don’t blow off steps)
Others said that this AI seems to be very helpful in regulating children’s basketball.
Another smart person suggested that Brother AP design another Oscar-winning error detection AI.（Doge）
So having said that, how did this AI referee have such a “sparkling eye”?
Mainly track the trajectory of the ball and the number of steps
As you may have guessed, this AI referee is mainly based on computer vision（CV）created.
The AI mainly tracks two things: the trajectory of the ball and the number of steps a person takes.
For this purpose, the timing of the dribble will be detected first.
First, Brother AP wrote a set of CV algorithms to detect the bouncing of the ball and stream the camera’s view, that is, extract the video frames in sequence.
Then, AP created an Aegis v image color mask to identify and filter out the color of the basketball.
In the computer daemon, the basketball appears white, and all other objects of a different color than the basketball appear black; because only the ball is recognized and placed in the mask.
However, the interfering colors caused some irregular pixel blocks to appear around the basketball. In order to optimize this problem, AP brother deleted some post-processing codes and made a circle around the ball to make it look more regular.
In order to better track the bounce of the basketball, Brother AP measured its radius, obtained a vertex based on the radius of the ball and the center point, and then used a parabolic function to represent the trajectory of the ball. When the vertex reaches the minimum value, the basketball has touched the ground.
After the basketball identification problem is solved, the next step is to count the number of steps people take while dribbling the ball.
At first, Brother AP thought that using the off-the-shelf step counter on the Apple Watch would work, but it turned out that he was too naive-
The pedometer on the Apple Watch doesn’t update in real time.
So AP brother simply did it himself and designed a real-time counting pedometer.
He created an Android application that detects and counts steps based on acceleration in the three axes of x, y, and z, and feeds this data back to a Python program.
Finally, by combining the two datasets monitoring the trajectory of the ball and the number of steps, it is possible to determine whether a person has walking behavior.
However, at present this AI referee is not “perfect” enough. Some netizens pointed out that this AI does not seem to recognize the gather step that is popular in the NBA today.(Harden should be more familiar with this)。
In this regard, AP brother said that he will add this function to it later.
Are you optimistic about this AI basketball referee? By the way, AP brother has open sourced his algorithm on GitHub, interested partners can go and see~
Portal:
https://github.com/ayushpai/Basketball-Detector
Reference link:
