distance_pr的算子很快
使用opencv模仿实现一下
halcon 的 region使用rle编码，还有可能使用凸包优化，simd，二分查找，多线程计算，这里只实现基础的功能
#include <opencv2/opencv.hpp>
#include <vector>
#include <limits>
#include <algorithm>// 结构体表示RLE编码的区域点
struct RLEPoint {int y;int x_start;int x_end;
};// 从二值图像生成RLE编码的区域表示
std::vector<RLEPoint> binaryImageToRLE(const cv::Mat& binImage) {std::vector<RLEPoint> rlePoints;CV_Assert(binImage.type() == CV_8UC1);for (int y = 0; y < binImage.rows; ++y) {const uchar* row = binImage.ptr<uchar>(y);bool inRun = false;int runStart = 0;for (int x = 0; x < binImage.cols; ++x) {if (row[x] > 0) { // 前景像素if (!inRun) {inRun = true;runStart = x;}} else {if (inRun) {inRun = false;rlePoints.push_back({y, runStart, x-1});}}}if (inRun) {rlePoints.push_back({y, runStart, binImage.cols-1});}}return rlePoints;
}// 计算凸包点集
std::vector<cv::Point> computeConvexHull(const std::vector<RLEPoint>& rlePoints) {std::vector<cv::Point> allPoints;// 将RLE点转换为离散点for (const auto& rle : rlePoints) {for (int x = rle.x_start; x <= rle.x_end; ++x) {allPoints.emplace_back(x, rle.y);}}// 计算凸包std::vector<cv::Point> hull;cv::convexHull(allPoints, hull);return hull;
}// 射线法判断点是否在凸包内
bool isPointInConvexHull(const cv::Point& p, const std::vector<cv::Point>& hull) {if (hull.size() < 3) return false;// 检查点是否在所有边的同一侧int prevSide = 0;for (size_t i = 0; i < hull.size(); ++i) {const cv::Point& p1 = hull[i];const cv::Point& p2 = hull[(i+1)%hull.size()];// 计算叉积int cross = (p2.x - p1.x) * (p.y - p1.y) - (p2.y - p1.y) * (p.x - p1.x);if (cross == 0) continue; // 在边上int currentSide = cross > 0 ? 1 : -1;if (prevSide == 0) {prevSide = currentSide;} else if (currentSide != prevSide) {return false;}}return true;
}// 优化的distance_pr实现
double optimized_distance_pr(const cv::Mat& region, const cv::Point& queryPoint) {// 1. 转换为RLE编码表示auto rlePoints = binaryImageToRLE(region);if (rlePoints.empty()) return std::numeric_limits<double>::max();// 2. 计算凸包auto convexHull = computeConvexHull(rlePoints);// 3. 检查查询点是否在凸包内bool inside = isPointInConvexHull(queryPoint, convexHull);// 4. 计算最小距离double minDist = std::numeric_limits<double>::max();if (inside) {// 如果在凸包内，距离可能为0，需要检查精确包含关系for (const auto& rle : rlePoints) {if (rle.y == queryPoint.y) {if (queryPoint.x >= rle.x_start && queryPoint.x <= rle.x_end) {return 0.0; // 点在区域内}}}}// 5. 使用凸包顶点快速估算搜索范围double searchRadius = std::numeric_limits<double>::max();for (const auto& p : convexHull) {double dx = p.x - queryPoint.x;double dy = p.y - queryPoint.y;double dist = std::sqrt(dx*dx + dy*dy);if (dist < searchRadius) {searchRadius = dist;}}// 6. 在RLE点集中搜索最小距离for (const auto& rle : rlePoints) {// 快速检查是否可能在搜索范围内if (std::abs(rle.y - queryPoint.y) > searchRadius) continue;// 计算到该行的最小距离int closestX = std::max(rle.x_start, std::min(rle.x_end, queryPoint.x));double dx = closestX - queryPoint.x;double dy = rle.y - queryPoint.y;double distance = std::sqrt(dx*dx + dy*dy);if (distance < minDist) {minDist = distance;// 更新搜索半径if (distance < searchRadius) {searchRadius = distance;}}}return minDist;
}
