BOE_ZB_BACKPLATE/AlgorithmModule/src/AI_Edge_Algin.cpp

/*
 * @Author: xiewenji 527774126@qq.com
 * @Date: 2025-09-11 15:32:52
 * @LastEditors: xiewenji 527774126@qq.com
 * @LastEditTime: 2025-09-22 16:56:41
 * @FilePath: /BOE_ZB_PLATE_Detect/AlgorithmModule/src/AI_Edge_Algin.cpp
 * @Description: 这是默认设置,请设置`customMade`, 打开koroFileHeader查看配置 进行设置: https://github.com/OBKoro1/koro1FileHeader/wiki/%E9%85%8D%E7%BD%AE
 */

#include "AI_Edge_Algin.h"
#include "CheckErrorCodeDefine.hpp"

#define EDGE_GPU 0

#include <opencv2/opencv.hpp>  // OpenCV核心功能
#include <opencv2/imgproc.hpp> // 图像处理模块
#include <vector>              // 向量容器
#include <iostream>            // 输入输出流

/**
 * @class SimilarityTransform
 * @brief 用于计算和存储相似变换（旋转、缩放、平移）
 *
 * 通过两个点对计算模板图像到检测图像的相似变换
 */
class SimilarityTransform
{
public:
    /**
     * @brief 构造函数，使用两个点对初始化变换矩阵
     *
     * @param template_p1 模板点1
     * @param template_p2 模板点2
     * @param detected_d1 检测点1
     * @param detected_d2 检测点2
     */
    SimilarityTransform(const cv::Point2f &template_p1,
                        const cv::Point2f &template_p2,
                        const cv::Point2f &detected_d1,
                        const cv::Point2f &detected_d2)
    {
        std::vector<cv::Point2f> template_pts = {template_p1, template_p2};
        std::vector<cv::Point2f> detected_pts = {detected_d1, detected_d2};

        // 计算相似变换矩阵
        transform_matrix_ = cv::estimateAffinePartial2D(template_pts, detected_pts);

        if (transform_matrix_.empty())
        {
            valid_ = false;
            std::cerr << "Error: Failed to calculate transformation matrix." << std::endl;
        }
        else
        {
            valid_ = true;
            extractParameters();
        }
    }

    /**
     * @brief 检查变换矩阵是否有效
     * @return bool 变换是否有效
     */
    bool isValid() const { return valid_; }

    /**
     * @brief 获取旋转角度（度）
     * @return double 旋转角度（度）
     */
    double getRotationAngle() const { return rotation_angle_; }

    /**
     * @brief 获取缩放比例
     * @return double 缩放比例
     */
    double getScaleFactor() const { return scale_factor_; }

    /**
     * @brief 获取平移分量
     * @return cv::Point2f 平移向量(tx, ty)
     */
    cv::Point2f getTranslation() const
    {
        return cv::Point2f(transform_matrix_.at<double>(0, 2),
                           transform_matrix_.at<double>(1, 2));
    }

    /**
     * @brief 转换点坐标
     *
     * @param point 输入点（模板坐标系）
     * @return cv::Point2f 转换后的点（检测图像坐标系）
     */
    cv::Point2f transformPoint(const cv::Point2f &point) const
    {
        if (!valid_)
        {
            std::cerr << "Warning: Using invalid transform! Returning original point." << std::endl;
            return point;
        }

        // 使用矩阵乘法进行点变换
        cv::Mat point_mat = (cv::Mat_<double>(3, 1) << point.x, point.y, 1);
        cv::Mat result_mat = transform_matrix_ * point_mat;

        return cv::Point2f(result_mat.at<double>(0), result_mat.at<double>(1));
    }

    /**
     * @brief 批量转换点坐标
     *
     * @param points 输入点集（模板坐标系）
     * @return std::vector<cv::Point2f> 转换后的点集（检测图像坐标系）
     */
    std::vector<cv::Point2f> transformPoints(const std::vector<cv::Point2f> &points) const
    {
        if (!valid_)
        {
            std::cerr << "Warning: Using invalid transform! Returning original points." << std::endl;
            return points;
        }

        std::vector<cv::Point2f> result;
        cv::transform(points, result, transform_matrix_);
        return result;
    }

    /**
     * @brief 获取变换矩阵
     * @return cv::Mat 2x3变换矩阵
     */
    cv::Mat getTransformMatrix() const { return transform_matrix_; }

private:
    /**
     * @brief 从变换矩阵中提取旋转角度和缩放比例
     */
    void extractParameters()
    {
        double a = transform_matrix_.at<double>(0, 0);
        double b = transform_matrix_.at<double>(0, 1);

        // 计算旋转角度（弧度转角度）
        rotation_angle_ = std::atan2(b, a) * 180.0 / CV_PI;

        // 计算缩放比例
        scale_factor_ = std::sqrt(a * a + b * b);
    }

    cv::Mat transform_matrix_;  // 2x3 变换矩阵
    bool valid_ = false;        // 变换是否有效
    double rotation_angle_ = 0; // 旋转角度（度）
    double scale_factor_ = 1;   // 缩放比例
};

/**
 * @brief 独立转换函数（使用变换矩阵）
 *
 * @param point 输入点（模板坐标系）
 * @param transform_matrix 2x3变换矩阵
 * @return cv::Point2f 转换后的点（检测图像坐标系）
 */
cv::Point2f transformPoint(const cv::Point2f &point, const cv::Mat &transform_matrix)
{
    // 验证变换矩阵有效性
    if (transform_matrix.empty() || transform_matrix.rows != 2 || transform_matrix.cols != 3)
    {
        std::cerr << "Error: Invalid transformation matrix! Returning original point." << std::endl;
        return point;
    }

    // 使用矩阵乘法进行点变换
    cv::Mat point_mat = (cv::Mat_<double>(3, 1) << point.x, point.y, 1);
    cv::Mat result_mat = transform_matrix * point_mat;

    return cv::Point2f(result_mat.at<double>(0), result_mat.at<double>(1));
}

AI_Edge_Algin::AI_Edge_Algin(std::shared_ptr<DetLog> &log_ref)
    : m_pdetlog(log_ref)
{
    m_bInitialized = false;
    m_bModelSucc = false;
    m_bModel_Mark_Succ = false;
    m_bshowimg = false;
    m_strRootPath_Big = "/home/aidlux/BOE/Algin/";
    m_strRootPath_Mark = "/home/aidlux/BOE/MarkLine/";
    creatsavedir();
    std::string m_strSavePath;

    m_pImageStorage = ImageStorage::getInstance();
}

AI_Edge_Algin::~AI_Edge_Algin()
{
}

int AI_Edge_Algin::Init(OtherDet_Config *pOtherDet_Config)
{
    m_pOtherDet_Config = pOtherDet_Config;

    AI_Factory = AIFactory::GetInstance();

    m_bInitialized = true;
    return 0;
}
int AI_Edge_Algin::Detect(const cv::Mat &img, DetConfig *pDetConfig, std::shared_ptr<Edge_AI_Result> &pCheckResult_Aling)
{
    m_pCheckResult_Aling = std::make_shared<Edge_AI_Result>();
    pCheckResult_Aling = m_pCheckResult_Aling;
    m_pdetlog->AddCheckstr(PrintLevel_2, DET_LOG_LEVEL_3, "AI_Edge_Algin ", "start");

    m_pDetConfig = pDetConfig;
    static int erridx = 0;
    std::string str_error = "";

    // 保存过程图片
    if (m_pDetConfig->IsSaveProcessImg())
    {
        erridx++;
        if (erridx > 9999999)
        {
            erridx = 0;
        }
        str_error = "/home/aidlux/BOE/Edge/Error/" + std::to_string(erridx) + "_src.png";
    }

    if (img.empty())
    {
        return 1;
    }

    // 1、初步定位 找到产品大致区域
    int re = 0;

    cv::RotatedRect Big_roi;
    cv::RotatedRect Small_roi;

    re = Get_Edge(0, img, pDetConfig, m_pDetConfig->strChannel, Big_roi);
    if (re != 0)
    {
        m_pdetlog->AddCheckstr(PrintLevel_2, DET_LOG_LEVEL_3, "AI_Edge_Algin ", "AICheck_Edge_Big----error  %d ", re);
        if (m_pDetConfig->IsSaveProcessImg())
        {
            cv::imwrite(str_error, img);
        }
        return re;
    }
    re = Get_Edge(1, img, pDetConfig, m_pDetConfig->strChannel, Small_roi);
    if (re != 0)
    {
        m_pdetlog->AddCheckstr(PrintLevel_2, DET_LOG_LEVEL_3, "AI_Edge_Algin ", "AICheck_Edge_Small----error  %d ", re);
        if (m_pDetConfig->IsSaveProcessImg())
        {
            cv::imwrite(str_error, img);
        }
        return re;
    }

    m_pCheckResult_Aling->bigroi = Big_roi;
    m_pCheckResult_Aling->smallroi = Small_roi;

    pCheckResult_Aling = m_pCheckResult_Aling;
    // return 1;

    return 0;
}

int AI_Edge_Algin::SaveSmallImg(const cv::Mat &img, const cv::Mat &mask, cv::Rect roi)
{
    // 是否要保存中间过程的小图
    if (m_pDetConfig->IsSaveProcessImg())
    {
        static int svsmallidx = 0;
        svsmallidx++;
        if (svsmallidx > 9999999)
        {
            svsmallidx = 0;
            /* code */
        }
        bool bssss = false;

        if (m_pDetConfig->saveProcessImg == Save_Filter)
        {
            // 4. 查找轮廓
            vector<vector<Point>> contours;
            vector<Vec4i> hierarchy;
            findContours(mask.clone(), contours, hierarchy, RETR_EXTERNAL, CHAIN_APPROX_SIMPLE);
            if (contours.size() > 1)
            {
                bssss = true;
            }
            else if (contours.size() == 1)
            {
                int pointNum = 0;
                for (int i = 0; i < contours.size(); i++)
                {
                    pointNum += contours[i].size();
                }
                // printf("pointNum============== %d\n", pointNum);
                if (pointNum > 30)
                {
                    bssss = true;
                }
            }
        }
        else if (m_pDetConfig->saveProcessImg == Save_ALL)
        {
            bssss = true;
        }

        if (bssss)
        {
            std::string str1 = "/home/aidlux/BOE/Edge/Smasll/" + std::to_string(svsmallidx) + "_in.png";
            std::string str2 = "/home/aidlux/BOE/Edge/Smasll/" + std::to_string(svsmallidx) + "_in_mask.png";
            // std::string st3 = "/home/aidlux/BOE/Edge/Smasll/" + std::to_string(svsmallidx) + "_in_show.png";
            cv::imwrite(str1, img);
            cv::imwrite(str2, mask);
            // cv::Mat showsss = temDet + smask * 0.4;
            // cv::imwrite(st3, showsss);
        }
    }
    return 0;
}

int AI_Edge_Algin::InitModel_ALL()
{

    m_bModelSucc = true;
    return 0;
}

int AI_Edge_Algin::Get_Edge(int AIModel_type, const cv::Mat &img, DetConfig *pDetConfig, std::string strChannel, cv::RotatedRect &Roi)
{
    std::shared_ptr<AIModel_Base> pBackPlate_Align;
    switch (AIModel_type)
    {
    case 0:
        pBackPlate_Align = AI_Factory->Align_Outer;
        break;
    case 1:
        pBackPlate_Align = AI_Factory->Align_Inner;
        break;
    default:
        pBackPlate_Align = AI_Factory->Align_Outer;
        break;
    }
    cv::Size sz;
    sz.width = pBackPlate_Align->input_0.width;
    sz.height = pBackPlate_Align->input_0.height;
    cv::Mat detImg;
    cout<< pDetConfig->strChannel << ": " << "---Get_Edge-resize-" << to_string(AIModel_type) <<"--   ";
    cout << "imgSize: " << img.size() << ", " << "detImgSize: " << detImg.size() << ", " << "szSize: " << sz << endl;

    cv::resize(img, detImg, sz);
    int re = 0;
    cv::Mat mask;
    if (detImg.channels() != 1)
    {
        cv::cvtColor(detImg, detImg, cv::COLOR_RGB2GRAY);
    }

    re = pBackPlate_Align->AIDet(detImg, mask);
    if (re != 0)
    {
        m_pdetlog->AddCheckstr(PrintLevel_3, DET_LOG_LEVEL_3, "AI_Edge_Algin ", "AICheck_Edge_%d----error  %d ", AIModel_type, re);
        int re123 = 100 + re;
        return re123;
    }

    if (pDetConfig->pBaseCheckFunction->saveImg.bSaveAlginImg)
    {
        creatsavedir();
        static int sdk = 0;
        std::string str = m_strSavePath_Big + std::to_string(sdk) + "_" + pDetConfig->strChannel + "_Img.png";
        int sr = m_pImageStorage->addImage(str, detImg);
        if (sr == 0)
        {
            str = m_strSavePath_Big + std::to_string(sdk) + "_" + pDetConfig->strChannel + "_Img_mask.png";
            int sr = m_pImageStorage->addImage(str, mask, true);
            sdk++;
            if (sdk > 99999)
            {
                sdk = 0;
                /* code */
            }
        }
    }

    // if (m_pDetConfig->bSaveResultImg)
    {
        cv::imwrite(strChannel +"_edge_"+ to_string(AIModel_type) +"_in.png", detImg);
        cv::imwrite(strChannel +"_edge_"+ to_string(AIModel_type) +"_out_mask.png", mask);
    }

    // 找到最大轮廓
    bool found;
    std::vector<cv::Point> boundingBox = CheckUtil::getLargestContourROI(mask, found);
    if (!found)
    {
        m_pdetlog->AddCheckstr(PrintLevel_3, DET_LOG_LEVEL_3, "AI_Edge_Algin ", "No contours found!----error ");
        return 4;
    }

    // 边缘轮廓点集还原为原图坐标，并计算得到RotateROI
    if (sz.width <= 0 || sz.height <= 0)
    {
        m_pdetlog->AddCheckstr(PrintLevel_3, DET_LOG_LEVEL_3, "AI_Edge_Algin ", "invalid model input size!----error ");
        return 5;
    }

    // 检测小图 到 原图 的缩放比例
    const float scale_x = static_cast<float>(img.cols) / sz.width;
    const float scale_y = static_cast<float>(img.rows) / sz.height;

    // 轮廓点还原到原图坐标系
    std::vector<cv::Point2f> srcContour;
    srcContour.reserve(boundingBox.size());
    for (const auto &pt : boundingBox)
    {
        srcContour.emplace_back(pt.x * scale_x, pt.y * scale_y);
    }

    if (srcContour.size() < 3)
    {
        m_pdetlog->AddCheckstr(PrintLevel_3, DET_LOG_LEVEL_3, "AI_Edge_Algin ", "contour points too few!----error ");
        return 6;
    }

    Roi = cv::minAreaRect(srcContour);

    return 0;
}


int AI_Edge_Algin::creatsavedir()
{
    std::string curDate = CheckUtil::getCurrentDate();
    if (curDate == m_strLastDate)
    {
        return 0;
    }
    m_strLastDate = curDate;
    m_strSavePath_Big = m_strRootPath_Big + curDate + "/";
    m_strSavePath_Mark = m_strRootPath_Mark + curDate + "/";

    CheckUtil::CreateDir(m_strSavePath_Big);
    CheckUtil::CreateDir(m_strSavePath_Mark);

    return 0;
    return 0;
}

Image_Feature_Algin::Image_Feature_Algin()
{
}

Image_Feature_Algin::~Image_Feature_Algin()
{
}

int Image_Feature_Algin::Detect(DetConfig *pDetConfig, Align_Result *pResult, std::vector<std::string> &LogList)
{
    // 检测目标：找到 参数模版图到 检测图的 映射关系。包括 缩放和移动。

    // 先缩放 在 移动
    std::string strlog = "";
    if (!pDetConfig)
    {
        return 1;
    }
    if (pDetConfig->TemplateImg.empty())
    {
        strlog = m_PrintLog.printstr(Print_Level_Error, "Image_Align", "TemplateImg is empty ");
        LogList.push_back(strlog);
        return 1;
    }
    // 裁切位置；
    pResult->Crop_Roi_DetImg = pDetConfig->DetImg_CropROi;
    // 1、缩放尺度
    float fx = 1;
    float fy = 1;
    // 裁切尺寸 存在 并合理
    if (pDetConfig->param_CropRoi.width > 0 && pDetConfig->param_CropRoi.height > 0 &&
        pDetConfig->DetImg_CropROi.width > 0 && pDetConfig->DetImg_CropROi.height > 0)
    {
        fx = pDetConfig->DetImg_CropROi.width * 1.0f / pDetConfig->param_CropRoi.width;
        fy = pDetConfig->DetImg_CropROi.height * 1.0f / pDetConfig->param_CropRoi.height;
        if (fx > 0.5 && fx < 2 && fy > 0.5 && fy < 2)
        {
            pResult->fCropROI_Scale_ParmToDet_X = fx;
            pResult->fCropROI_Scale_ParmToDet_Y = fy;
        }
        else
        {
            strlog = m_PrintLog.printstr(Print_Level_Error, "Image_Align", "Scale out  0.5--2");
            LogList.push_back(strlog);
            return 1;
        }
    }
    else
    {
        strlog = m_PrintLog.printstr(Print_Level_Error, "Image_Align", "crop ROI Error");
        LogList.push_back(strlog);
        return 1;
    }
    // strlog = m_PrintLog.printstr(Print_Level_Info, "Image_Align", "Scale x %f y %f\n", fx, fy);
    // LogList.push_back(strlog);

    // 2、定位
    // 1)、模版特征图片的 缩放。
    cv::Mat TemplateFeature;
    cv::Size sz;
    // fx = 1;
    // fy = 1;
    sz.width = int(pDetConfig->TemplateImg.cols * fx);
    sz.height = int(pDetConfig->TemplateImg.rows * fy);
    cv::resize(pDetConfig->TemplateImg, TemplateFeature, sz);

    if (!CheckUtil::RoiInImg(pDetConfig->Search_Roi, pDetConfig->DetImg))
    {
        strlog = m_PrintLog.printstr(Print_Level_Error, "Image_Align", "Search_Roi ROI Error Not In img");
        LogList.push_back(strlog);
        return 1;
    }

    cv::Mat DetFeature = pDetConfig->DetImg(pDetConfig->Search_Roi).clone();

    double confidence = 0;

    int kernel_size = 128;
    int search_size = 1024;
    int det_search_min_size = DetFeature.cols;
    if (DetFeature.rows < det_search_min_size)
    {
        det_search_min_size = DetFeature.rows;
    }
    int template_kernel_min_size = TemplateFeature.cols;
    if (TemplateFeature.rows < template_kernel_min_size)
    {
        template_kernel_min_size = TemplateFeature.rows;
    }
    float f_search = search_size * 1.0f / det_search_min_size;
    float f_Kernel = kernel_size * 1.0f / template_kernel_min_size;
    float falign = f_search;
    if (f_Kernel > falign)
    {
        falign = f_Kernel;
    }
    cv::Size Search_sz;
    Search_sz.width = int(DetFeature.cols * falign);
    Search_sz.height = int(DetFeature.rows * falign);
    cv::Mat Search_img;
    cv::resize(DetFeature, Search_img, Search_sz);
    cv::Size Kernel_sz;
    Kernel_sz.width = int(TemplateFeature.cols * falign);
    Kernel_sz.height = int(TemplateFeature.rows * falign);
    cv::Mat Kernel_img;
    cv::resize(TemplateFeature, Kernel_img, Kernel_sz);

    auto bestMatch = findBestTemplateMatch(Search_img, Kernel_img, confidence);

    bestMatch.x /= falign;
    bestMatch.y /= falign;

    pResult->bestMatch = bestMatch;
    if (pDetConfig->bSaveImg)
    {
        cv::imwrite("Align_template.png", TemplateFeature);
        cv::imwrite("Align_Det.png", DetFeature);
    }

    if (confidence != -1)
    {
        std::cout << "最佳匹配位置: (" << bestMatch.x << ", " << bestMatch.y
                  << "), 得分: " << confidence << std::endl;
        if (confidence > pDetConfig->fscore)
        {

            /* code */

            int m_x = pDetConfig->feature_Roi.x * fx - pDetConfig->Search_Roi.x;
            int m_y = pDetConfig->feature_Roi.y * fy - pDetConfig->Search_Roi.y;
            pResult->offt_x = bestMatch.x - m_x;
            pResult->offt_y = bestMatch.y - m_y;
            // printf("m_x %d bestMatch.x %d offt_x %d\n", m_x, bestMatch.x, pResult->offt_x);
            // printf("m_y %d bestMatch.y %d offt_y %d\n", m_y, bestMatch.y, pResult->offt_y);

            pResult->bDet = true;

            pResult->Crop_Roi_ParmImg = pResult->Det_srcToParm_src_Rect(pResult->Crop_Roi_DetImg);

            strlog = m_PrintLog.printstr(Print_Level_Info, "Image_Align", "  -- Succ :Align score  %f > %f offt x %d y %d Scale x %f y %f",
                                         confidence, pDetConfig->fscore, pResult->offt_x, pResult->offt_y, pResult->fCropROI_Scale_ParmToDet_X, pResult->fCropROI_Scale_ParmToDet_Y);

            LogList.push_back(strlog);
        }
        else
        {

            strlog = m_PrintLog.printstr(Print_Level_Error, "Image_Align", "  error :Align score  %f< 0.9", confidence);
            pResult->fCropROI_Scale_ParmToDet_X = 1;
            pResult->fCropROI_Scale_ParmToDet_Y = 1;
            LogList.push_back(strlog);
        }
    }
    else
    {
        pResult->fCropROI_Scale_ParmToDet_X = 1;
        pResult->fCropROI_Scale_ParmToDet_Y = 1;
        std::cout << "未找到有效匹配" << std::endl;
    }

    return 0;
}

cv::Point Image_Feature_Algin::findBestTemplateMatch(
    const cv::Mat &detectionImage,
    const cv::Mat &templateImage,
    double &bestScore,
    int method)
{
    // 输入验证
    if (detectionImage.empty() || templateImage.empty())
    {
        throw std::invalid_argument("输入图像不能为空");
    }
    if (detectionImage.channels() != 1 || templateImage.channels() != 1)
    {
        throw std::invalid_argument("必须输入灰度图像");
    }
    if (templateImage.rows > detectionImage.rows ||
        templateImage.cols > detectionImage.cols)
    {
        throw std::invalid_argument("模板尺寸不能大于被检测图像");
    }

    // cv::imwrite("detectionImage.png", detectionImage);
    // cv::imwrite("templateImage.png", templateImage);
    // 执行模板匹配
    cv::Mat resultMatrix;
    cv::matchTemplate(detectionImage, templateImage, resultMatrix, method);

    // 确定极值搜索方式
    const bool findMinima = (method == cv::TM_SQDIFF || method == cv::TM_SQDIFF_NORMED);

    // 查找极值位置
    cv::Point extremaLoc = cv::Point(0, 0);
    double extremaVal;
    cv::minMaxLoc(resultMatrix,
                  findMinima ? &extremaVal : nullptr,
                  findMinima ? nullptr : &extremaVal,
                  findMinima ? &extremaLoc : nullptr,
                  findMinima ? nullptr : &extremaLoc);

    // 设置有效性检查阈值（可根据方法动态调整）
    double threshold = 0.0;
    switch (method)
    {
    case cv::TM_CCOEFF_NORMED:
        threshold = 0.6;
        break; // [-1, 1]
    case cv::TM_CCORR_NORMED:
        threshold = 0.7;
        break; // [0, 1]
    case cv::TM_SQDIFF_NORMED:
        threshold = 0.2;
        break; // [0, 1]
    default:
        threshold = 0.0;
    }

    // 验证匹配有效性
    const bool isValid = findMinima ? (extremaVal <= threshold) : (extremaVal >= threshold);

    if (isValid)
    {
        bestScore = extremaVal;
        return extremaLoc;
    }
    bestScore = -1; // 无效时的默认值
    return extremaLoc;
}