Implementing a WeChat Jump Game Bot Using Python and OpenCV

WeChat recently introduced the "Jump" mini‑game, which quickly became popular; many developers created bots that automatically play the game by capturing the screen, calculating the distance to the next platform, and simulating a press via ADB. This article describes the computer‑vision approach using Python and OpenCV to identify the player and the landing point.

Player detection is performed with template matching because the player character has a fixed purple shape. After saving a screenshot of the character as player.png and the game screen as 1.png , the following code finds the best match:

import cv2 as cv
img = cv.imread("1.png")
player_template = cv.imread('player.png')
player = cv.matchTemplate(img, player_template, cv.TM_CCOEFF_NORMED)
min_val, max_val, min_loc, max_loc = cv.minMaxLoc(player)

The coordinates max_loc give the top‑left corner of the player. To visualize the result, the code draws a rectangle around the player and a circle at its centre:

corner_loc = (max_loc[0] + 50, max_loc[1] + 150)
player_spot = (max_loc[0] + 25, max_loc[1] + 150)
cv.circle(img, player_spot, 10, (0, 255, 255), -1)
cv.rectangle(img, max_loc, corner_loc, (0, 0, 255), 5)
cv.namedWindow('img', cv.WINDOW_KEEPRATIO)
cv.imshow("img", img)
cv.waitKey(0)

Landing‑point detection relies on edge detection because the platforms have various shapes. First the image is blurred to reduce noise and then the Canny operator extracts edges:

img_blur = cv.GaussianBlur(img, (5, 5), 0)  # Gaussian blur
canny_img = cv.Canny(img_blur, 1, 10)        # Canny edge detection
cv.namedWindow('img', cv.WINDOW_KEEPRATIO)
cv.imshow("img", canny_img)

Only the upper half of the screen contains useful information, so the image is cropped to remove the score board and the lower part:

height, width = canny_img.shape
crop_img = canny_img[300:int(height/2), 0:width]
cv.namedWindow('img', cv.WINDOW_KEEPRATIO)
cv.imshow("img", crop_img)

Because a small portion of the player’s head may still appear in the cropped area, those pixels are cleared to avoid interference:

for y in range(max_loc[1], max_loc[1]+150):
    for x in range(max_loc[0], max_loc[0]+50):
        canny_img[y][x] = 0

To find the landing platform’s centre, the algorithm scans the cropped edge image for the first white pixel (point A) and the pixel with the maximum x‑coordinate (point B). The centre is derived from these two points:

crop_h, crop_w = crop_img.shape
center_x, center_y = 0, 0
max_x = 0
for y in range(crop_h):
    for x in range(crop_w):
        if crop_img[y, x] == 255:
            if center_x == 0:
                center_x = x
            if x > max_x:
                center_y = y
                max_x = x
cv.circle(crop_img, (center_x, center_y), 10, 255, -1)
cv.namedWindow('img', cv.WINDOW_KEEPRATIO)
cv.imshow("img", crop_img)
cv.waitKey(0)

With the player position and the landing‑point centre known, the Euclidean distance between them can be computed, and the required press duration is derived from this distance. The article notes that the same result could also be achieved with findContours but the direct pixel‑scan method is simpler for this case.

All source code is available on the author’s GitHub repository: https://github.com/Yigang0622/OpenCV_Wechat_jump .