iOS中基于OpenCV实现物体尺寸测量及宽高计算的问题

Hey there! Let's work through this object dimension measurement problem with OpenCV on iOS. I see you've got the CvVideoCamera feed up and running, but hit a snag in the processImage method. Let's break this down step by step to get you measuring object widths and heights properly.

First, let's finish and refine your edge processing code #

You got started with grayscale conversion, Gaussian blur, and Canny edge detection, but left off mid-morphological operations. Here's how to complete that foundation, plus add the critical steps for contour detection and dimension calculation:

- (void)processImage:(cv::Mat &)image {
    // 1. Convert BGRA camera feed to grayscale (reduces computation load)
    cv::Mat gray;
    // Note: Use COLOR_BGRA2GRAY for OpenCV 3+, CV_BGRA2GRAY for older versions
    cvtColor(image, gray, cv::COLOR_BGRA2GRAY);

    // 2. Gaussian blur to reduce noise before edge detection
    cv::GaussianBlur(gray, gray, cv::Size(7, 7), 0);

    // Optional: Add histogram equalization to improve contrast in low-light scenarios
    cv::equalizeHist(gray, gray);

    // 3. Canny edge detection to isolate object boundaries
    cv::Canny(gray, gray, 50, 100);

    // 4. Morphological operations: Dilate then erode to eliminate small noise blobs
    cv::Mat kernel = cv::getStructuringElement(cv::MORPH_RECT, cv::Size(3, 3));
    cv::dilate(gray, gray, kernel);
    cv::erode(gray, gray, kernel);

    // 5. Find contours of objects in the processed image
    std::vector<std::vector<cv::Point>> contours;
    std::vector<cv::Vec4i> hierarchy;
    // Retrieve only outer contours to ignore nested noise
    cv::findContours(gray, contours, hierarchy, cv::RETR_EXTERNAL, cv::CHAIN_APPROX_SIMPLE);

    // 6. Filter and process valid contours
    for (size_t i = 0; i < contours.size(); i++) {
        double contourArea = cv::contourArea(contours[i]);
        // Skip tiny contours (adjust threshold based on your camera resolution)
        if (contourArea < 500) continue;

        // 7. Get the bounding rectangle for the contour
        cv::Rect objectRect = cv::boundingRect(contours[i]);

        // 8. Draw the rectangle on the original image for visualization
        cv::rectangle(image, objectRect, cv::Scalar(0, 255, 0), 2);

        // 9. Calculate real-world dimensions (CRITICAL: Need calibration first!)
        // --- Calibration Step ---
        // Use a reference object with known dimensions (e.g., a credit card: 85.6mm wide)
        // Measure its pixel width in the feed first to get the conversion ratio
        const float referenceRealWidthMM = 85.6; // Known width of your reference object
        const float referencePixelWidth = 300.0; // Pixel width of reference in your feed (calibrate this!)
        float mmPerPixel = referenceRealWidthMM / referencePixelWidth;

        // Convert pixel dimensions to real-world millimeters
        float realWidthMM = objectRect.width * mmPerPixel;
        float realHeightMM = objectRect.height * mmPerPixel;

        // 10. Draw the dimension text on the image
        std::string dimensionText = cv::format("W: %.1fmm, H: %.1fmm", realWidthMM, realHeightMM);
        cv::putText(image, dimensionText, cv::Point(objectRect.x, objectRect.y - 10),
                    cv::FONT_HERSHEY_SIMPLEX, 0.5, cv::Scalar(0, 255, 0), 2);
    }
}

Key Notes for Success #

• Calibration is non-negotiable: The conversion ratio (mmPerPixel) depends on the distance between the camera and the object. Always calibrate with a known-size object placed at the same distance as your target objects first.
• OpenCV version checks: If you're using an older OpenCV version (pre-3.0), replace COLOR_BGRA2GRAY with CV_BGRA2GRAY.
• Better object filtering: For rectangular objects, you can use cv::approxPolyDP to check if the contour has 4 vertices (a rectangle), which will reduce false positives from irregular shapes.
• Camera mirroring: By default, CvVideoCamera mirrors front-facing camera feeds. If this throws off your visualization, set camera->mirrorFrontFacing = YES; (or mirrorRearFacing) to adjust.

内容的提问来源于stack exchange，提问作者Dharma