基于OpenCV&C++(ObjC++)的正方形检测：角点提取与验证求助

Hey there! You’re already halfway there by extracting the largest region—nice work. Let’s walk through how to grab those four corners and validate if the shape is a square, since you hit a snag at this stage.

Step 1: Refine the Contour & Get 4 Vertices #

First, your extracted region might have minor irregularities or small protrusions. We’ll use a convex hull to smooth out the contour, then apply polygon approximation to narrow it down to exactly 4 vertices (the square’s corners):

// Assume you already have the largest contour stored in 'largestContour'
std::vector<cv::Point> hull;
cv::convexHull(largestContour, hull);

// Approximate the convex hull to a simplified polygon
std::vector<cv::Point> approx;
double epsilon = 0.02 * cv::arcLength(hull, true); // Adjust based on your image scale
cv::approxPolyDP(hull, approx, epsilon, true);

// Ensure we get a quadrilateral (4 points)
if (approx.size() != 4) {
    // Handle non-quadrilateral case: tweak epsilon or clean up the contour first
    return;
}

Step 2: Sort the Corners (Left-Top, Right-Top, Left-Bottom, Right-Bottom) #

Now we need to order the 4 points so we can consistently reference each corner. Here’s a straightforward sorting method based on coordinate values:

// Helper function to sort corners into standard order
void sortCorners(std::vector<cv::Point>& corners) {
    // First sort by x-coordinate (left to right)
    std::sort(corners.begin(), corners.end(), [](const cv::Point& a, const cv::Point& b) {
        return a.x < b.x;
    });

    // Split into left and right pairs, then sort each by y-coordinate (top to bottom)
    std::vector<cv::Point> leftPair = {corners[0], corners[1]};
    std::vector<cv::Point> rightPair = {corners[2], corners[3]};
    
    std::sort(leftPair.begin(), leftPair.end(), [](const cv::Point& a, const cv::Point& b) {
        return a.y < b.y;
    });
    std::sort(rightPair.begin(), rightPair.end(), [](const cv::Point& a, const cv::Point& b) {
        return a.y < b.y;
    });

    // Reassign to standard order: left-top, right-top, left-bottom, right-bottom
    corners[0] = leftPair[0];
    corners[1] = rightPair[0];
    corners[2] = leftPair[1];
    corners[3] = rightPair[1];
}

// Call the sorter on your approximated points
sortCorners(approx);

Step 3: Validate if It’s a Square #

You mentioned checking side lengths or angles—let’s cover both. Note: Squares have 90° angles, so I’ll assume the 45° mention was a typo, but I’ll include a note if you meant something else like a rhombus with 45° angles.

// Calculate all four side lengths using Euclidean distance
std::vector<double> sideLengths;
sideLengths.push_back(cv::norm(approx[0] - approx[1])); // Top side
sideLengths.push_back(cv::norm(approx[1] - approx[3])); // Right side
sideLengths.push_back(cv::norm(approx[3] - approx[2])); // Bottom side
sideLengths.push_back(cv::norm(approx[2] - approx[0])); // Left side

// Check if all sides are roughly equal (allow 10% tolerance—adjust for your use case)
double avgLength = std::accumulate(sideLengths.begin(), sideLengths.end(), 0.0) / 4;
bool sidesMatch = true;
for (double len : sideLengths) {
    if (std::abs(len - avgLength) > 0.1 * avgLength) {
        sidesMatch = false;
        break;
    }
}

// Check if angles are ~90° using dot product
cv::Point vecTop = approx[1] - approx[0];
cv::Point vecLeft = approx[2] - approx[0];
double dotProduct = vecTop.x * vecLeft.x + vecTop.y * vecLeft.y;
double angle = std::acos(dotProduct / (cv::norm(vecTop) * cv::norm(vecLeft))) * 180 / CV_PI;

bool anglesValid = (std::abs(angle - 90) < 5); // Allow 5° tolerance

// Final check for square
if (sidesMatch && anglesValid) {
    std::cout << "Success! This is a square." << std::endl;
} else {
    std::cout << "Not a square—tweak tolerance or clean up the contour further." << std::endl;
}

Quick Troubleshooting Tips #

• Epsilon issues: If approxPolyDP isn’t returning 4 points, adjust the epsilon value (try 0.01 to 0.05 * arcLength).
• Noisy contours: Run a cv::morphologyEx with a closing kernel first to fill small holes or smooth edges.
• Tolerance tweaks: If your images are low-quality, loosen the side length tolerance (e.g., 15%) or angle tolerance (e.g., 8°).

内容的提问来源于stack exchange，提问作者user9590073