Skip to content

A Computer-Vision-based project to hack Audition game

Notifications You must be signed in to change notification settings

WildanEriono/auto-audition

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

33 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Auto Audition Game Hack Software

Auto play Audition game by using basic image processing techniques.

This project was done in 2 days for the purpose of killing my time.

Commercializing this project is prohibited and illegal.

Results

Example 1
Example 2

Installation

  1. Create environment
conda create --name au python=3.7.7
conda activate au
conda install -c conda-forge jupyterlab
pip install opencv-python
pip install pillow
pip install imutils
pip install keyboard
pip install mss
  1. Clone example-app.conf and rename to app.conf

  2. Find your game PID in Task Manager and modify the pid item in app.conf file.

  3. Run app.py file with admin permission.

    • Press F5 button if you want to recompute the speed.
    • Press F6 or F7 button if you want to adjust the perfect time.
    • Press backspace to quit program

How it works?

Firstly, the game window will be scanned by using python-mss.

Secondly, the region of keys (arrows) and the perfect area (the area which has the running marker) will be cropped. These regions are in fixed positions.

Perfect area
Keys area

Finally, some tricky image processing algorithms will be performed on the keys area to detect the keys. Simultaneously, some algorithms will also be applied in the perfect area to detect the time when we should hit Ctrl (to hit perfect).

Key detection

From the keys area image, I threshold it to get the arrows only. Then the contour detection algorithm will help us to find those separated arrows. I sort the contours to make it sequential from left to right. Now I have each key separated as a ROI (region of interest) by getting the bounding boxes of those contours.

Thresholded keys
Contours

For each ROI (bounding box of key's contour) as below, I split it into 4 regions red, blue, green, and purple. The region that has the highest amount of white pixels will contain the arrow's head.

Keys detection

For the blue/red key, I convert the original RGB image into HSV color space then just apply this trick to detect whether this ROI is blue-ish or red-ish.

Perfect detection

There are 2 main tasks here. The first one is to detect the marker position. The other one is to measure the time we need to hit Ctrl.

For the first task, I use template matching algorithm to measure the correlation coefficients of the 2 images which are the perfect area and the marker itself. The matchest position is the marker's position.

Template matching

By getting 2 marker's positions at 2 consecutive periods, I interpolate the speed of the marker.

For the second task, we just get the current marker's position, then calculate the time it takes to go to the Perfect Position by using the above computed speed.

Perfect position

Combination

I create 2 threads. One thread is to control the keys. The other one is to control when we should hit Ctrl.

I also define "one loop" is the period of time that the marker runs from the head to the end (tail) or the perfect bar.

One loop

On the keys controlling thread, I wait until the marker is at the head part. Then I perform keys detection and send the keyboard code to the game window. Then I wait until the marker is at the tail part. That is to make sure the keys detection will be performed one time per loop.

For the Ctrl hitting thread, the Ctrl hitting time detection will be started only when the marker passed the middle of the perfect bar. This is to make sure the algorithm doesn't measure the hitting time too soon. The farther the marker is to the perfect position, the worse the measurement will be.

Head middle tail

Because the measurement of the time we should hit Ctrl is not good enough, every time it calculates I will add/subtract a small amount of time. I use Page Up and Page Down buttons to adjust that small changes. keyboard library was used to listen to the global key events.

Keyboard keys are sent to the game window by using KeyboardCtrl class (check keyboard_ctrl.py).

Limitation

I don't design any mechanism to get feedback on how accurate the measurement is. Therefore, there's no way to adjust the hitting Ctrl time.

The project should be done by using the Reinforcement-Learning-based method or some PID-based algorithm with the feedback mechanism to adjust the control.

Contributing

Any contribution is welcome.

About

A Computer-Vision-based project to hack Audition game

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Jupyter Notebook 57.7%
  • Python 34.8%
  • C++ 7.5%