识别部分

由于底层逻辑是识别黑胶带轮廓来确定A4纸的位置, 而题目描述中有提到一条宽为5mm黑色基准线, 当A4纸目标物紧贴地面, 即紧贴基准线时, 黑胶带外轮廓会与基准线重合, 这时可能就会导致无法正确识别A4纸目标物, 所以我们识别黑胶带的内轮廓来确定A4纸的位置。

所以先定义黑胶带内轮廓的数据

1
2
A4_WIDTH_MM = 170
A4_HEIGHT_MM = 257

识别黑胶带内轮廓

首先我们需要对输入图像进行预处理：

  1. 转换为灰度图像
  2. 应用高斯模糊
  3. 使用边缘检测算法

接着使用cv2.findContours(edged, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)来寻找所有轮廓并存放进valid_contours数组中以便后续采用第二大的轮廓作为Target的内轮廓

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
def find_a4_contour(image):
    """寻找矩形框，总计两个矩形框"""`
    roi_x, roi_y, roi_w, roi_h = (526, 222, 227, 276)
    roi_image = image[roi_y:roi_y+roi_h, roi_x:roi_x+roi_w]

    cv2.rectangle(image, (roi_x, roi_y), (roi_x + roi_w, roi_y + roi_h), (255, 0, 0), 2)

    gray = cv2.cvtColor(roi_image, cv2.COLOR_BGR2GRAY)
    blur = cv2.GaussianBlur(gray, (5,5), 0)
    edged=cv2.threshold(blur, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)[1]
    contours, _ = cv2.findContours(edged, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
    #cv2.imshow("Edges", edged)
    
    # 先筛选出四边形轮廓
    valid_contours = []
    for cnt in contours:
        area = cv2.contourArea(cnt)
        if area > 5000:  # 只保留四边形
            valid_contours.append(cnt)
            #print(f"Contour Area: {area}, Approx Length: {len(approx_1)}")
    
    # 对筛选后的四边形轮廓进行排序
    if len(valid_contours) < 2:
        return None
    sorted_contours = sorted(valid_contours, key=cv2.contourArea, reverse=True)
    target = sorted_contours[1]
    # cv2.drawContours(roi_image, [target], -1, (0, 255, 0), 2)  # 绘制目标轮廓
    # cv2.drawContours(roi_image, [sorted_contours[0]], -1, (255, 255, 0), 2)  # 绘制最大轮廓
    # cv2.imshow("Target Contour", roi_image)  # 显示目标轮廓
    
    #对target轮廓进行多边形逼近
    epsilon = 0.02 * cv2.arcLength(target, True)
    approx = cv2.approxPolyDP(target, epsilon, True)
    # cv2.imshow("Approx Contour", roi_image)  # 显示逼近后的轮廓       
    if len(approx) == 4:
        x, y, w, h = cv2.boundingRect(approx)
        aspect_ratio = float(w) / h
        if 0.35 < aspect_ratio < 0.75: #筛选掉过宽或过窄的轮廓
            # 对target进行透视变换
            warped = get_topdown_view(roi_image, order_points(approx.reshape(4, 2)))
            #显示变换后的宽高
            cv2.putText(image, f"W: {warped.shape[1]}, H: {warped.shape[0]}", (roi_x, roi_y - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)
            # cv2.imshow("Topdown View", warped)
            cv2.drawContours(roi_image, [approx], -1, (0, 255, 0), 2)
            return warped
    return None

透视变换

首先需要对识别到的矩形轮廓进行角点排序, 以便后续的透视变换

1
2
3
4
5
6
7
8
9
10
11
def order_points(pts):
    """对A4角点排序为：左上，右上，右下，左下"""
    rect = np.zeros((4, 2), dtype="float32")
    s = pts.sum(axis=1)
    diff = np.diff(pts, axis=1)

    rect[0] = pts[np.argmin(s)]     # 左上
    rect[2] = pts[np.argmax(s)]     # 右下
    rect[1] = pts[np.argmin(diff)]  # 右上
    rect[3] = pts[np.argmax(diff)]  # 左下
    return rect

当A4纸水平旋转时, 在镜头看起来, A4纸的宽会随着旋转的角度产生明显的偏差, 而A4纸的高只会因为两个边距离镜头的距离不同产生细微的变化。所以我们可以通过已知的A4纸比例和测算的像素高推算出矫正后的像素宽应当为多少

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
def get_topdown_view(image, pts):
    """透视变换,将A4纸变为标准正视图，保持原始像素尺寸"""
    # 计算A4纸在图像中的实际像素尺寸
    width_top = np.linalg.norm(pts[1] - pts[0])      # 上边长度
    width_bottom = np.linalg.norm(pts[2] - pts[3])   # 下边长度
    height_left = np.linalg.norm(pts[3] - pts[0])    # 左边长度
    height_right = np.linalg.norm(pts[2] - pts[1])   # 右边长度
    
    #对高取平均值，宽通过A4纸比例计算
    max_height = int((height_left + height_right) / 2)
    max_width = int(max_height * (A4_WIDTH_MM / A4_HEIGHT_MM))

    # 目标点坐标，保持检测到的实际像素尺寸
    dst = np.array([
        [0, 0],
        [max_width, 0],
        [max_width, max_height],
        [0, max_height]
    ], dtype="float32")

    M = cv2.getPerspectiveTransform(pts, dst)
    warped = cv2.warpPerspective(image, M, (max_width, max_height))
    return warped