一个图像测距的问题，需要用到图像校正。
萌新小白求大神解答，功能主要是实现跳远距离的测量，请问怎么校正这个透视产生的梯形，目前能想到的就是插值的方法，通过插值来校正成矩形，但不知道具体怎么实施，直接检测沙坑四个点做透视变换的误差在3cm左右，效果还不理想，求大神们解答。
由于你的功能是测量跳远距离，那么主要在意纵向精度，那么你在纵向位置选择9个点，得到9个点的图像坐标，然后去测量这9个点，得到这9个点的实际距离，找一个图像算法库里面的方法生成标定文件。 有了标定文件就可以把你的任意图像坐标转换为实际距离（同样图像算法库里有现成的方法）。亲测9点标定在3个像素精度
你可能需要看看世界坐标系，相机坐标系，图像坐标系这几个改变。
先阈值分割 检测沙坑形成的梯形，然后将整幅图做透视变换还原为真实场景，再通过沙坑的长宽实际距离来计算距离
很有意思，这是刚找的四个点：
# 寻找四点
def findTrapezoidal(oImg):
font = cv2.FONT_HERSHEY_SIMPLEX
# mImg = imutils.resize(oImg, width=1000) #缩小
img = oImg.copy()
# 色调
hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
sat = hsv[:, :, 1]
# threshold and invert 阈值和反转
thresh = cv2.threshold(sat, 10, 255, cv2.THRESH_BINARY)[1]
thresh = 255 - thresh
# apply morphology dilate 形态扩张
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (15, 15))
thresh = cv2.morphologyEx(thresh, cv2.MORPH_DILATE, kernel)
# contour 轮廓
cnts, hierarchy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_NONE)
points = []
for cnt in cnts:
area = cv2.contourArea(cnt)
# 找最大的：这是内框
if 5_800_000 > area > 5_400_000:
print('area: ', area)
approx = cv2.approxPolyDP(cnt, 0.009 * cv2.arcLength(cnt, True), True)
# 画框
cv2.drawContours(img, cnt, -1, (0, 255, 75), 5)
# 找 vertices 顶点
n = approx.ravel()
i = 0
for j in n:
if i % 2 == 0:
x = n[i]
y = n[i + 1]
# String containing the co-ordinates.
string = str(x) + " " + str(y)
if i == 0:
# 最高的
# print(f'顶点: {x}, {y}')
points.append((x, y))
# cv2.putText(img, "Arrow tip", (x, y-5),
# font, 0.5, (255, 0, 0))
else:
# 剩余的
# print(string)
points.append((x, y))
# cv2.putText(img, string, (x, y-5),
# font, 0.5, (0, 255, 0))
i = i + 1
points = sorted(points, key=lambda k: [k[1], k[0]])
print('找到的点： ', points)
if len(points) == 4:
print(f'上右： {points[0]}, 上左：{points[1]}\t\t下右： {points[3]}, 下左：{points[2]}', points)
for pt in points:
cv2.circle(img, pt, 5, [0, 0, 255], 100)
return img
# 坐标排列
def order_points(pts):
# 坐标启动
rect = np.zeros((4, 2), dtype="float32")
# 左上 有最小的 和
# 右下 有最大的 和
s = np.sum(pts, axis=1)
rect[0] = pts[np.argmin(s)]
rect[2] = pts[np.argmax(s)]
# 右上 有最小的 差
# 左下 有最大的 差
diff = np.diff(pts, axis=1)
rect[1] = pts[np.argmin(diff)]
rect[3] = pts[np.argmax(diff)]
return rect
# 鸟视转换
def four_point_transform(pts, img):
# 坐标排列
rect = order_points(pts)
(tl, tr, br, bl) = rect
# 计算 最高宽度（x 轴）：右下 --》 左下 对比 右上 --》 左上
widthA = np.sqrt(((br[0] - bl[0]) ** 2) + ((br[1] - bl[1]) ** 2))
widthB = np.sqrt(((tr[0] - tl[0]) ** 2) + ((tr[1] - tl[1]) ** 2))
maxWidth = max(int(widthA), int(widthB))
# 计算 最高高度（y 轴）：右上 --》 右下 对比 左上 --》 左下
heightA = np.sqrt(((tr[0] - br[0]) ** 2) + ((tr[1] - br[1]) ** 2))
heightB = np.sqrt(((tl[0] - bl[0]) ** 2) + ((tl[1] - bl[1]) ** 2))
maxHeight = max(int(heightA), int(heightB))
# 鸟视
dst = np.array([
[0, 0],
[maxWidth - 1, 0],
[maxWidth - 1, maxHeight - 1],
[0, maxHeight - 1]], dtype="float32")
M = cv2.getPerspectiveTransform(rect, dst)
warped = cv2.warpPerspective(img, M, (maxWidth, maxHeight))
return warped
好像转错了。
改改：
# 寻找四点
def findTrapezoidal(oImg):
font = cv2.FONT_HERSHEY_SIMPLEX
# mImg = imutils.resize(oImg, width=1000) #缩小
img = oImg.copy()
# 色调
hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
sat = hsv[:, :, 1]
# threshold and invert 阈值和反转
thresh = cv2.threshold(sat, 10, 255, cv2.THRESH_BINARY)[1]
thresh = 255 - thresh
# apply morphology dilate 形态扩张
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (15, 15))
thresh = cv2.morphologyEx(thresh, cv2.MORPH_DILATE, kernel)
# contour 轮廓
cnts, hierarchy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_NONE)
points = []
for cnt in cnts:
area = cv2.contourArea(cnt)
# 找最大的：这是内框
if 5_800_000 > area > 5_400_000:
print('area: ', area)
approx = cv2.approxPolyDP(cnt, 0.009 * cv2.arcLength(cnt, True), True)
# 画框
cv2.drawContours(img, cnt, -1, (0, 255, 75), 5)
# 找 vertices 顶点
n = approx.ravel()
i = 0
for j in n:
if i % 2 == 0:
x = n[i]
y = n[i + 1]
# String containing the co-ordinates.
string = str(x) + " " + str(y)
if i == 0:
# 最高的
# print(f'顶点: {x}, {y}')
points.append((x, y))
# cv2.putText(img, "Arrow tip", (x, y-5),
# font, 0.5, (255, 0, 0))
else:
# 剩余的
# print(string)
points.append((x, y))
# cv2.putText(img, string, (x, y-5),
# font, 0.5, (0, 255, 0))
i = i + 1
print('找到的点： ', points)
if len(points) == 4:
print(f'右上： {points[0]}, 右下：{points[1]}\t\t左上： {points[3]}, 左下：{points[2]}')
cv2.circle(img, points[0], 5, [0, 0, 255], 100) # Red
cv2.circle(img, points[1], 5, [205, 0, 0], 100) # Blue
cv2.circle(img, points[2], 5, [139, 0, 139], 100) # Magenta
cv2.circle(img, points[3], 5, [47, 255, 173], 100) # Green
return points, img
然后对号改改：
# 坐标排列
def order_points(pts):
# 坐标启动
rect = np.zeros((4, 2), dtype="float32")
# 左上 有最小的 和
# 右下 有最大的 和
s = np.sum(pts, axis=1)
rect[1] = pts[np.argmin(s)]
rect[3] = pts[np.argmax(s)]
# 右上 有最小的 差
# 左下 有最大的 差
diff = np.diff(pts, axis=1)
rect[2] = pts[np.argmin(diff)]
rect[0] = pts[np.argmax(diff)]
return rect
鸟视更新：
# 鸟视转换
def four_point_transform(pts, img):
# 坐标排列
# print('old: ', pts)
rect = order_points(pts)
# print('new: ', rect)
(tl, tr, br, bl) = rect
# 计算 最高宽度（x 轴）：右下 --》 左下 对比 右上 --》 左上
widthA = np.sqrt(((br[0] - bl[0]) ** 2) + ((br[1] - bl[1]) ** 2))
widthB = np.sqrt(((tr[0] - tl[0]) ** 2) + ((tr[1] - tl[1]) ** 2))
maxWidth = max(int(widthA), int(widthB))
print('maxW: ', maxWidth)
# 计算 最高高度（y 轴）：右上 --》 右下 对比 左上 --》 左下
heightA = np.sqrt(((tr[0] - br[0]) ** 2) + ((tr[1] - br[1]) ** 2))
heightB = np.sqrt(((tl[0] - bl[0]) ** 2) + ((tl[1] - bl[1]) ** 2))
maxHeight = max(int(heightA), int(heightB))
print('maxH: ', maxHeight)
# 鸟视
dst = np.array([
[0, 0],
[maxHeight - 1, 0],
[maxHeight - 1, maxWidth - 1],
[0, maxWidth - 1]], dtype="float32")
M = cv2.getPerspectiveTransform(rect, dst)
# result = cv2.warpPerspective(frame, matrix, (500, 600))
warped = cv2.warpPerspective(img, M, (maxHeight, maxWidth))
return warped缩小比例：
# 更新图像尺寸
def ResizeWithAspectRatio(image, width=None, height=None, inter=cv2.INTER_AREA):
dim = None
(h, w) = image.shape[:2]
if width is None and height is None:
return image
if width is None:
r = height / float(h)
dim = (int(w * r), height)
else:
r = width / float(w)
dim = (width, int(h * r))
return cv2.resize(image, dim, interpolation=inter)显示：
# =====================================
# main
# =====================================
# load the example image
image = cv2.imread(".\\test\\sandbox.jpg")
corners, imgCp = findTrapezoidal(image)
shiftImg = four_point_transform(corners, imgCp)
# image = cv2.rotate(shiftImg, cv2.cv2.ROTATE_90_CLOCKWISE)
imgResized = ResizeWithAspectRatio(imgCp, width=400)
cpResized = ResizeWithAspectRatio(shiftImg, height=400, width=500)
# =====================================
# Window 窗口： Demo
# =====================================
demo_win = 'Demo'
cv2.namedWindow(demo_win)
cv2.moveWindow(demo_win, 400, 80)
cv2.resizeWindow(demo_win, 400, 300)
# =====================================
# Window 窗口： Selected
# =====================================
select_win = 'Selected'
cv2.namedWindow(select_win)
cv2.moveWindow(select_win, 810, 80)
# cv2.resizeWindow(select_win, 500, 400)
while True:
cv2.imshow(demo_win, imgResized)
cv2.imshow(select_win, cpResized)
# keyboard control
key = cv2.waitKey(1)
if key == ord("q") or key == ord("Q"):
break
cv2.destroyAllWindows()这次翻对了：挺好玩的，就没有尺寸，无法标注。
9点标定，或者使用标定板标定，标定完毕后转换图像坐标，通过转换的坐标来求距离， 4点仿射变换精度<9点标定精度<标定板标定精度