Skip to content

Commit

Permalink
otpmize tic-tac-toe
Browse files Browse the repository at this point in the history
  • Loading branch information
@Neutree
Neutree committed Jul 30, 2024
1 parent cba4436 commit 6b5ae12
Showing 1 changed file with 131 additions and 93 deletions.
224 changes: 131 additions & 93 deletions projects/demo_tic_tac_toe/main.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -5,9 +5,11 @@
'''


from types import CellType
from maix import camera, display, image, time, app, touchscreen
import cv2
import numpy as np
from collections import deque

# 80 fps or 60fps
cam = camera.Camera(320, 320, fps=80)
Expand Down Expand Up @@ -35,25 +37,60 @@ def key_clicked(btn_rects):
def check_mode_switch(img : image.Image, disp_w, disp_h):
global mode
btns = [
[0, 0, 100, 40],
[0, 0, 75, 30],
]
btn_rects_disp = [image.resize_map_pos(img.width(), img.height(), disp_w, disp_h, image.Fit.FIT_CONTAIN, btns[0][0], btns[0][1], btns[0][2], btns[0][3])]
clicked, idx, rect = key_clicked(btn_rects_disp)
if clicked:
mode += 1
if mode > 2:
if mode > 3:
mode = 1
img.draw_string(2, 10, f"mode {mode}", color=image.COLOR_WHITE, scale=1.5)
img.draw_string(2, 5, f"mode{mode}", color=image.COLOR_WHITE, scale=1.3)
img.draw_rect(btns[0][0], btns[0][1], btns[0][2], btns[0][3], image.COLOR_WHITE, 2)

def find_grid_centers(corners):
"""
corners: List of 4 tuples representing the corners of the 3x3 grid in the order:
top-left, top-right, bottom-right, bottom-left
"""
# Define the 3x3 grid points in the normalized coordinate space
grid_points = np.array([[0, 0], [1/3, 0], [2/3, 0], [1, 0],
[0, 1/3], [1/3, 1/3], [2/3, 1/3], [1, 1/3],
[0, 2/3], [1/3, 2/3], [2/3, 2/3], [1, 2/3],
[0, 1], [1/3, 1], [2/3, 1], [1, 1]])

# Convert corners to numpy array
src_points = np.array(corners, dtype=np.float32)

# Define destination points for the perspective transform
dst_points = np.array([[0, 0], [1, 0], [1, 1], [0, 1]], dtype=np.float32)

# Compute the perspective transform matrix
transform_matrix = cv2.getPerspectiveTransform(dst_points, src_points)

# Transform grid points to image coordinate space
transformed_points = cv2.perspectiveTransform(np.array([grid_points], dtype=np.float32), transform_matrix)[0]

# Calculate the center points of the grid cells
centers = []
for i in range(0, 3):
for j in range(0, 3):
cx = (transformed_points[i * 4 + j][0] + transformed_points[i * 4 + j + 1][0] +
transformed_points[(i + 1) * 4 + j][0] + transformed_points[(i + 1) * 4 + j + 1][0]) / 4
cy = (transformed_points[i * 4 + j][1] + transformed_points[i * 4 + j + 1][1] +
transformed_points[(i + 1) * 4 + j][1] + transformed_points[(i + 1) * 4 + j + 1][1]) / 4
centers.append((int(cx), int(cy)))

return centers

def find_qipan():
find_center_method = 1 # 1, 2
area_threshold = 80
pixels_threshold = 50
thresholds = [[0, 80, 0, 80, 30, 80]] # red
# thresholds = [[0, 80, -120, -10, 0, 30]] # green
# thresholds = [[0, 80, 30, 100, -120, -60]] # blue
while mode == 1:
while mode == 1 or mode == 2:
find_center_method = mode # 1：根据4个角确定格子中心, 2：使用找色块确定格子中心
img = cam.read()
check_mode_switch(img, disp.width(), disp.height())

Expand All @@ -64,18 +101,29 @@ def find_qipan():
gray = cv2.cvtColor(img_cv, cv2.COLOR_RGB2GRAY)

# 高斯模糊去噪声
# blurred = cv2.GaussianBlur(gray, (5, 5), 0)
blurred = cv2.GaussianBlur(gray, (5, 5), 0)

# 边缘检测
edged = cv2.Canny(gray, 50, 150)
# 边缘检测，阈值 0，150
edged = cv2.Canny(blurred, 50, 150)

# 膨胀处理
kernel = np.ones((5, 5), np.uint8)
dilated = cv2.dilate(edged, kernel, iterations=1)

# disp.show(image.cv2image(dilated, False, False))
# 腐蚀处理
eroded = cv2.erode(dilated, kernel, iterations=1)

binary = eroded

# 显示二值化后的图
# 左下角显示
# img_tmp = image.cv2image(binary, False, False).resize(img.width() // 4, img.height() // 4)
# img.draw_image(0, img.height() - img_tmp.height(), img_tmp)
# 整张图显示
# disp.show(image.cv2image(binary, False, False))

# 查找轮廓
contours, _ = cv2.findContours(dilated, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
contours, _ = cv2.findContours(binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
if len(contours) > 0:
# 筛选出最大的轮廓
largest_contour = max(contours, key=cv2.contourArea)
Expand All @@ -88,107 +136,97 @@ def find_qipan():
if len(approx) == 4:
# print("rect found")
corners = approx.reshape((4, 2))
# 按顺序排列角点（左上、右上、右下、左下）
rect = np.zeros((4, 2), dtype="int")
s = corners.sum(axis=1)
rect[0] = corners[np.argmin(s)]
rect[2] = corners[np.argmax(s)]
diff = np.diff(corners, axis=1)
rect[1] = corners[np.argmin(diff)]
rect[3] = corners[np.argmax(diff)]
corners = rect

# TODO: 这是外角点，可以优化一下找到线中心的点，这里直接写死往内收拢 1 个像素
half_x = 1
half_y = 1
corners[0] += (half_x, half_y)
corners[1] += (-half_x, half_y)
corners[2] += (-half_x, -half_y)
corners[3] += (half_x, -half_y)

# 绘制顶点和轮廓
# for corner in corners:
# cv2.circle(img_cv, tuple(corner), 4, (0, 255, 0), -1)
for corner in corners:
cv2.circle(img_cv, tuple(corner), 4, (0, 255, 0), -1)

# 绘制四边路径性
# cv2.drawContours(img_cv, [approx], -1, (0, 255, 0), 1)

# 洪泛填充外部，如果棋盘外部的背景和棋盘内部的背景不同才需要这一步
img.flood_fill(corners[0][0] - 5, corners[0][1] - 5, 0.3, 0.3, image.COLOR_BLUE)
# img.flood_fill(corners[1][0] - 5, corners[1][1] + 5, 0.5, 0.05, image.COLOR_BLUE)
# img.flood_fill(corners[0][0] + 5, corners[0][1] + 5, 0.5, 0.05, image.COLOR_BLUE)
# img.flood_fill(corners[0][0] + 5, corners[0][1] - 5, 0.5, 0.05, image.COLOR_BLUE)

# 按顺序排列角点（左上、右上、右下、左下）
# rect = np.zeros((4, 2), dtype="float32")
# s = corners.sum(axis=1)
# rect[0] = corners[np.argmin(s)]
# rect[2] = corners[np.argmax(s)]
# diff = np.diff(corners, axis=1)
# rect[1] = corners[np.argmin(diff)]
# rect[3] = corners[np.argmax(diff)]

# (tl, tr, br, bl) = rect
# 上面出来的结果已经是 点从左上逆时针，所以跳过找顶点
tl = corners[0]
bl = corners[1]
br = corners[2]
tr = corners[3]

# 计算3x3格子的交叉点
cross_points = []
for i in range(4):
for j in range(4):
# 线性插值计算交叉点
cross_x = int((tl[0] * (3 - i) + tr[0] * i) * (3 - j) / 9 +
(bl[0] * (3 - i) + br[0] * i) * j / 9)
cross_y = int((tl[1] * (3 - i) + tr[1] * i) * (3 - j) / 9 +
(bl[1] * (3 - i) + br[1] * i) * j / 9)
cross_points.append((cross_x, cross_y))
cv2.circle(img_cv, (cross_x, cross_y), 3, (0, 255, 0), -1)
# 洪泛填充外部，如果棋盘外部的背景和棋盘内部的背景相同且后面用了找色块的方式找才需要这一步
if find_center_method == 2:
img.flood_fill(corners[0][0] - 5, corners[0][1] - 5, 0.3, 0.3, image.COLOR_BLUE)
# img.flood_fill(corners[1][0] - 5, corners[1][1] + 5, 0.5, 0.05, image.COLOR_BLUE)
# img.flood_fill(corners[0][0] + 5, corners[0][1] + 5, 0.5, 0.05, image.COLOR_BLUE)
# img.flood_fill(corners[0][0] + 5, corners[0][1] - 5, 0.5, 0.05, image.COLOR_BLUE)

centers = []
# 找格子中心点方法一：直接根据顶点计算
if find_center_method == 1:
for i in range(3):
for j in range(3):
center_x = int((cross_points[i * 4 + j][0] + cross_points[i * 4 + j + 1][0] + cross_points[(i + 1) * 4 + j][0] + cross_points[(i + 1) * 4 + j + 1][0]) / 4)
center_y = int((cross_points[i * 4 + j][1] + cross_points[i * 4 + j + 1][1] + cross_points[(i + 1) * 4 + j][1] + cross_points[(i + 1) * 4 + j + 1][1]) / 4)
centers.append((center_x, center_y))
cv2.circle(img_cv, (center_x, center_y), 2, (0, 255, 0), -1)
elif find_center_method == 2:
# 找格子中心点方法二： 找色块的方式来确定中心点
# 根据顶点找出中心点
centers = find_grid_centers(corners)
# 找格子中心点方法二： 找色块的方式来确定中心点
elif find_center_method == 2: # 上面出来的结果已经是 点从左上顺时针
roi = [corners[:,0].min(), corners[:,1].min(), corners[:,0].max() - corners[:,0].min(), corners[:,1].max() - corners[:,1].min()]
img.draw_rect(roi[0], roi[1], roi[2], roi[3], image.COLOR_WHITE)
blobs = img.find_blobs(thresholds, roi=roi, x_stride=2, y_stride=1, area_threshold=area_threshold, pixels_threshold=pixels_threshold)
for b in blobs:
centers.append((b.cx(), b.cy()))
img.draw_circle(b.cx(), b.cy(), 2, image.COLOR_WHITE, -1)
# 对找到的中心点进行编号, y + x 最大就是右下角，最小就是左上角， y-x 最大就是左下角，y-x 最小就是右上角，其它几个点根据在旁边两个点中间判断
if len(centers) == 9:
centers = np.array(centers)
rect = np.zeros((9, 2), dtype="int")
s = centers.sum(axis=1)
idx_0 = np.argmin(s)
idx_8 = np.argmax(s)
diff = np.diff(centers, axis=1)
idx_2 = np.argmin(diff)
idx_6 = np.argmax(diff)
rect[0] = centers[idx_0]
rect[2] = centers[idx_2]
rect[6] = centers[idx_6]
rect[8] = centers[idx_8]
# 其它点
calc_center = (rect[0] + rect[2] + rect[6] + rect[8]) / 4
mask = np.zeros(centers.shape[0], dtype=bool)
idxes = [1, 3, 4, 5, 7]
mask[idxes] = True
others = centers[mask]
idx_l = others[:,0].argmin()
idx_r = others[:,0].argmax()
idx_t = others[:,1].argmin()
idx_b = others[:,1].argmax()
found = np.array([idx_l, idx_r, idx_t, idx_b])
mask = np.isin(range(len(others)), found, invert=False)
idx_c = np.where(mask == False)[0]
if len(idx_c) == 1:
rect[1] = others[idx_t]
rect[3] = others[idx_l]
rect[4] = others[idx_c]
rect[5] = others[idx_r]
rect[7] = others[idx_b]
else:
# 等于 45度的情况
print("== 45 degree")
centers = rect
else:
raise Exception("find_center_method value error")

# 对找到的中心点进行编号, y + x 最大就是右下角，最小就是左上角， y-x 最大就是左下角，y-x 最小就是右上角，其它几个点根据在旁边两个点中间判断
# 画出中心点
# 写编号
if len(centers) == 9:
centers = np.array(centers)
rect = np.zeros((9, 2), dtype="float32")
s = centers.sum(axis=1)
idx_0 = np.argmin(s)
idx_8 = np.argmax(s)
diff = np.diff(centers, axis=1)
idx_2 = np.argmin(diff)
idx_6 = np.argmax(diff)
rect[0] = centers[idx_0]
rect[2] = centers[idx_2]
rect[6] = centers[idx_6]
rect[8] = centers[idx_8]
# 其它点
calc_center = (rect[0] + rect[2] + rect[6] + rect[8]) / 4
mask = np.zeros(centers.shape[0], dtype=bool)
idxes = [1, 3, 4, 5, 7]
mask[idxes] = True
others = centers[mask]
idx_l = others[:,0].argmin()
idx_r = others[:,0].argmax()
idx_t = others[:,1].argmin()
idx_b = others[:,1].argmax()
found = np.array([idx_l, idx_r, idx_t, idx_b])
mask = np.isin(range(len(others)), found, invert=False)
idx_c = np.where(mask == False)[0]
if len(idx_c) == 1:
rect[1] = others[idx_t]
rect[3] = others[idx_l]
rect[4] = others[idx_c]
rect[5] = others[idx_r]
rect[7] = others[idx_b]
# 写编号
for i in range(9):
img.draw_string(rect[i][0], rect[i][1], f"{i + 1}", image.COLOR_WHITE, scale=2, thickness=-1)
else:
# 大于 45度的情况
print("> 45 degree")
for i in range(9):
x, y = centers[i][0], centers[i][1]
cv2.circle(img_cv, (x, y), 2, (0, 255, 0), -1)
img.draw_string(x, y, f"{i + 1}", image.COLOR_WHITE, scale=2, thickness=-1)

disp.show(img)

Expand All @@ -203,7 +241,7 @@ def find_qizi():
thresholds.append(threshold_black)
thresholds.append(threshold_white)

while mode == 2:
while mode == 3:
img = cam.read()
check_mode_switch(img, disp.width(), disp.height())

Expand Down

0 comments on commit 6b5ae12

Please sign in to comment.