yolo roi binary box find light

src/rm_auto_aim/armor_yolo_detect/include/armor_yolo_detect/armor_yolo_detector.hpp

@@ -146,7 +146,7 @@ private:
   std::vector<fyt::auto_aim::Armor> processROIs(
       const cv::Mat& input,
       const cv::Mat& gray_img,
-      const cv::Mat& binary_img);
+      const std::vector<cv::Rect>& rois);
 
   std::unique_ptr<YoloTensorRT> tensorrt_;
   std::vector<YoloObject> yolo_objects_;

src/rm_auto_aim/armor_yolo_detect/src/armor_yolo_detector.cpp

@@ -68,14 +68,6 @@ std::vector<fyt::auto_aim::Armor> Detector::detect(const cv::Mat& input) {
   cv::Mat gray_img;
   cv::cvtColor(input, gray_img, cv::COLOR_BGR2GRAY);
 
-  cv::Mat full_binary_img;
-  cv::threshold(gray_img, full_binary_img, binary_thres, 255, cv::THRESH_BINARY);
-
-  // Store binary image for debug visualization
-  if (enable_debug) {
-    debug_binary_img = full_binary_img.clone();
-  }
-
   uint64_t current_frame = frame_counter_.fetch_add(1, std::memory_order_relaxed);
 
   // Update YOLO objects based on mode
@@ -102,19 +94,7 @@ std::vector<fyt::auto_aim::Armor> Detector::detect(const cv::Mat& input) {
     return armors_;
   }
 
-  // Process ROIs to detect armors
-  armors_ = processROIs(input, gray_img, full_binary_img);
-
-  return armors_;
-}
-
-std::vector<fyt::auto_aim::Armor> Detector::processROIs(
-    const cv::Mat& input,
-    const cv::Mat& gray_img,
-    const cv::Mat& binary_img) {
-  std::vector<fyt::auto_aim::Armor> result_armors;
-
-  // 1. Calculate all YOLO ROIs for filtering
+  // Calculate ROIs from YOLO objects first
   std::vector<cv::Rect> rois;
   for (const auto& yolo_obj : yolo_objects_) {
     cv::Rect roi = expandBBox(yolo_obj.rect, input.size(), params.roi_expand_pixel);
@@ -124,31 +104,50 @@ std::vector<fyt::auto_aim::Armor> Detector::processROIs(
     }
   }
 
-  // 2. Find lights on the FULL image natively (exact armor_detector logic)
-  cv::Rect full_roi(0, 0, input.cols, input.rows);
-  auto all_lights = findLightsInROI(input, gray_img, binary_img, full_roi);
-
-  // 3. Filter lights: only keep those inside YOLO ROIs
-  std::vector<fyt::auto_aim::Light> filtered_lights;
-  for (const auto& light : all_lights) {
-    bool in_roi = false;
-    for (const auto& roi : rois) {
-      if (roi.contains(light.center)) {
-        in_roi = true;
-        break;
-      }
-    }
-    if (in_roi) {
-      filtered_lights.push_back(light);
-    }
+  if (rois.empty()) {
+    return armors_;
   }
 
-  if (filtered_lights.empty()) {
+  // Store binary image for debug visualization (crop to first ROI)
+  if (enable_debug) {
+    const auto& roi = debug_rois_.front();
+    cv::Mat roi_gray = gray_img(roi);
+    cv::threshold(roi_gray, debug_binary_img, binary_thres, 255, cv::THRESH_BINARY);
+  }
+
+  // Process ROIs to detect armors (only in ROI regions)
+  armors_ = processROIs(input, gray_img, rois);
+
+  return armors_;
+}
+
+std::vector<fyt::auto_aim::Armor> Detector::processROIs(
+    const cv::Mat& input,
+    const cv::Mat& gray_img,
+    const std::vector<cv::Rect>& rois) {
+  std::vector<fyt::auto_aim::Armor> result_armors;
+
+  // Find lights in each ROI region (with local binary for each ROI)
+  std::vector<fyt::auto_aim::Light> all_lights;
+  for (const auto& roi : rois) {
+    // Crop ROI from gray image
+    cv::Mat roi_gray = gray_img(roi);
+
+    // Apply binary threshold within ROI
+    cv::Mat roi_binary;
+    cv::threshold(roi_gray, roi_binary, binary_thres, 255, cv::THRESH_BINARY);
+
+    // Find lights in this ROI
+    auto lights = findLightsInROI(input, gray_img, roi_binary, roi);
+    all_lights.insert(all_lights.end(), lights.begin(), lights.end());
+  }
+
+  if (all_lights.empty()) {
     return result_armors;
   }
 
-  // 4. Match lights to form armors natively in global coordinates
-  auto armors = matchLightsInROI(filtered_lights);
+  // 4. Match lights to form armors in global coordinates
+  auto armors = matchLightsInROI(all_lights);
 
   // 5. Extract numbers, classify and refine corners
   if (classifier != nullptr) {
@@ -243,17 +242,29 @@ std::vector<fyt::auto_aim::Light> Detector::findLightsInROI(
       light = fyt::auto_aim::Light(contour);
     }
 
+    // Transform light coordinates from ROI-local to global
+    light.center += cv::Point2f(roi.x, roi.y);
+    light.top += cv::Point2f(roi.x, roi.y);
+    light.bottom += cv::Point2f(roi.x, roi.y);
+    // Also update the RotatedRect center
+    light.center = light.center;  // Already updated above
+
     if (isLight(light)) {
       // Color detection within ROI
       int sum_r = 0;
       int sum_b = 0;
       int sample_count = 0;
 
+      // ROI offset for coordinate transformation
+      const int roi_x = roi.x;
+      const int roi_y = roi.y;
+
       if (light_params.use_fit_line && !points.empty()) {
         auto b_rect = cv::boundingRect(contour);
         for (const auto& point : points) {
-          const int x = point.x + b_rect.x;
-          const int y = point.y + b_rect.y;
+          // Add both b_rect offset (relative to ROI) and ROI offset (global)
+          const int x = point.x + b_rect.x + roi_x;
+          const int y = point.y + b_rect.y + roi_y;
           const auto& pixel = rgb_img.at<cv::Vec3b>(y, x);
           sum_b += pixel[0];
           sum_r += pixel[2];
@@ -261,7 +272,10 @@ std::vector<fyt::auto_aim::Light> Detector::findLightsInROI(
         }
       } else {
         for (const auto& point : contour) {
-          const auto& pixel = rgb_img.at<cv::Vec3b>(point.y, point.x);
+          // Add ROI offset for global coordinates
+          const int x = point.x + roi_x;
+          const int y = point.y + roi_y;
+          const auto& pixel = rgb_img.at<cv::Vec3b>(y, x);
           sum_b += pixel[0];
           sum_r += pixel[2];
           ++sample_count;