Super Hexagon image processing gamer plan

Since this game was finally released on PC and it's a perfect field for real time image processing experiments, due to distinctive patterns and never ending gameplay i will try to do an auto player (certain words are missed due to google positioning system). Initially I want to programacaly go for 10s in hexagonest mode, we will see how that will go.

Targets

1. Feed the game feed to OpenCV

I will use dxtory to get a feed from the game, it will be delayed but while experimenting we will se if it will be a huge issue, since in this mode you have quite alot of time to react programically speaking. If the delay will be too huge i have a capture card that will feed me image with 1 frame lag. capture cards are visible as additional cameras in the system

2. find the center and find it's x,y,z rotations x,y,z differentials

This is the essential step, due to unique pattern in the middle we can track the middle and the offsets, they will affect the entire alforithm, we will try to normalize the image to normal view

3. find the player triangle.

We need to draw an elipse arround the middle, scale it up untill we hit a pixel from the triangle.

4. find the tracks, derermine count of tracks

Here i'm still thinking. I have several ideas that I don't want to share now, I need to see if they will be optimal.

5. track the obstacles on the track

We will track the obstacles, they may be nonrectangular that's why we will use a line scanning algorithm on the track, we will have 6 binary array that will determine if the track is closed, we wil feed that information with position of the player to the neural network.

6. create and learn CUDA neural network or an genetic algorithm that will determine the direction of spin

In this part we have 3 choices - stay, move left, move right we need determine it depending on which path we are, how many dimensions there are, distance to change path and


Also this shows how hard our eyes and brain needs to work to actually play this game :) but this algorithm don't care about color changing due to binarization of images and rotations of the entire map, which our brain must compensate.

That's all for now, i will see you around.