Skip to content

README.md

Latest commit

 

History

History
58 lines (33 loc) · 3.5 KB

README.md

File metadata and controls

58 lines (33 loc) · 3.5 KB

automatic_leap

play wechat game 'tiaoyitiao' automatically with python\opencv\adb shell\windows .bat

Illustrate in step with functions' name:

  1. start the game in android phone, and connect the phone with computer using USB, then make sure all the eviroment dependencies done.

GetScreenshot():

  1. use adb interface and windows .bat script to make shotscreen, then transfer the shotscreen fig into computer.

  2. pre-process the fig: To reduce CV computing time, only save the middle part of the fig into variables. For the fact that the game is built in 2 dims picture to show 3 dims space, and the angle of axis-x and axis-y is 120°, together with the limit of width of figuer pixel, only save middle part is OK.
    Save 3 channel of the picture into different variable to tackle with some situation that color change gradually.

  3. Also, to tackle with some situation that color change gradually.we analyze 5 pictures in a loop until one good result is found, they are b,g,r channel, and original 3-ch picture(use cv2.cvtColor() to transform it as a gray one, denoted as figure-B), and fig-B after equalizeHist process respectively.

PreProcess():

  1. preprocess the fig with Canny Arithmetic to find the edge， then dilate the edge for better detection in following steps.

LocateDollandMiniCircle():

  1. use cv2.matchTemplate() to find the jumping doll(character) and return its bottum center coordinate A (x1,y1) in picture.

    use cv2.matchTemplate() to find the mini circle in whole picture and return its center coordinate B (x2,y2).

    in continuous leap, if we always jump into the next bat's center part, then before next jump, the mini circle(next bat's center part) will appear.Continuously Jumping into this part will get a exponential growth in score.

  2. if template result is good in 4)& 5), we can calculate the next jump length. the next target pats only have two different shapes: 1. Rectangle(in xyz axis 3dim space, it is quad-angle like diamond) 2. circle (oval)

LocateRectangle(): return the center coordinates of Rectangle

  1. find the upper 120° angle in whole picture using cv2.matchTemplate(), since we know every jump direction is 30° and is in upper part of the doll, we can find the target pat if it is quad-angle.

LocateCircles(): return the center coordinates of circle

  1. resize the x-axis of the 2-dim picture, we can transfer oval into good circle, then we use cv2.HoughCircles() to detect target circle if possible.

  2. calculate the jump length in 7)& 8), compare them with the result of 6), get the max result.

MakeJmpAction():

  1. transfer the jump length into finger touch duration time, then use adb shell script to simulate human touch in phone.
  2. after a few second for waiting the doll finishing jump, back to 1) step for loop.