fix issues with calibration for large images

for larger cal images (~4kx4k) the keypoint finding routine wasn't working
well. Now it checks to be sure that the 3 keypoints are relatively the
same size (+/- 30%) to prevent the auto thresholding algorithm from
inadvertently setting the threshold based on three bogus keypoints

Author: dicengine

src/opencv/DICe_Calibration.cpp

@@ -567,6 +567,8 @@ Calibration::extract_dot_target_points(){
   const int_t orig_thresh_start = input_params_.get<int_t>(opencv_server_threshold_start,20);
   const int_t orig_thresh_end =   input_params_.get<int_t>(opencv_server_threshold_end,250);
   const int_t orig_thresh_step =  input_params_.get<int_t>(opencv_server_threshold_step,5);
+  if(!input_params_.isParameter(opencv_server_min_blob_size))
+    input_params_.set<int_t>(opencv_server_min_blob_size,100);
   scalar_t include_image_set_tol = 0.75; //the search must have found at least 75% of the total to be included
   for (size_t i_image = 0; i_image < num_images(); i_image++){
     //go through each of the camera's images (note only two camera calibration is currently supported)
@@ -590,13 +592,18 @@ Calibration::extract_dot_target_points(){
       int_t return_thresh = orig_thresh_start;
       int_t error_code = opencv_dot_targets(img, input_params_,
         key_points,img_points,grd_points,return_thresh);
+
+      // check to see if the min_blob_size needs to be adjusted, first try 10 (smaller), then try 500 (larger):
+      if(error_code!=0){
+      }
+
       // check if extraction failed
       if(error_code==0){
-        input_params_.set(opencv_server_threshold_start,return_thresh);
+        input_params_.set(opencv_server_threshold_start,return_thresh); // make all the values the same so the auto thresholding is skipped next time
         input_params_.set(opencv_server_threshold_end,return_thresh);
         input_params_.set(opencv_server_threshold_step,return_thresh);
       }else if(i_image!=0&&error_code==1){
-        input_params_.set(opencv_server_threshold_start,orig_thresh_start);
+        input_params_.set(opencv_server_threshold_start,orig_thresh_start); // reset the thresholds and re-run the auto thresholding routine
         input_params_.set(opencv_server_threshold_end,orig_thresh_end);
         input_params_.set(opencv_server_threshold_step,orig_thresh_step);
         error_code = opencv_dot_targets(img, input_params_,
@@ -605,6 +612,16 @@ Calibration::extract_dot_target_points(){
           input_params_.set(opencv_server_threshold_start,return_thresh);
           input_params_.set(opencv_server_threshold_end,return_thresh);
           input_params_.set(opencv_server_threshold_step,return_thresh);
+        }else{
+          input_params_.set<int_t>(opencv_server_min_blob_size,10);
+          DEBUG_MSG("Calibration::extract_dot_target_points(): resetting the min_blob_size to 10 and trying again.");
+          error_code = opencv_dot_targets(img, input_params_,
+            key_points,img_points,grd_points,return_thresh);
+          if(error_code==0){
+            input_params_.set(opencv_server_threshold_start,return_thresh);
+            input_params_.set(opencv_server_threshold_end,return_thresh);
+            input_params_.set(opencv_server_threshold_step,return_thresh);
+          }
         }
       }
 

src/opencv/DICe_OpenCVServerUtils.cpp

@@ -583,7 +583,20 @@ int_t opencv_dot_targets(cv::Mat & img,
   std::vector<cv::KeyPoint> key_points;
   std::vector<cv::KeyPoint> img_points;
   std::vector<cv::KeyPoint> grd_points;
-  return opencv_dot_targets(img,options,key_points,img_points,grd_points, return_thresh);
+  if(!options.isParameter(opencv_server_min_blob_size))
+    options.set<int_t>(opencv_server_min_blob_size,100);
+  int_t error_code = opencv_dot_targets(img,options,key_points,img_points,grd_points, return_thresh);
+  if(error_code!=0){
+    DEBUG_MSG("opencv_dot_targets(): resetting the min_blob_size to 10 and trying again.");
+    options.set<int_t>(opencv_server_min_blob_size,10);
+    error_code = opencv_dot_targets(img,options,key_points,img_points,grd_points, return_thresh);
+//    if(error_code!=0){
+//      DEBUG_MSG("opencv_dot_targets(): resetting the min_blob_size to 500 and trying again.");
+//      options.set<int_t>(opencv_server_min_blob_size,500);
+//      error_code = opencv_dot_targets(img,options,key_points,img_points,grd_points, return_thresh);
+//    }
+  }
+  return error_code;
 }
 
 DICE_LIB_DLL_EXPORT
@@ -602,7 +615,6 @@ int_t opencv_dot_targets(Mat & img,
   options.print(std::cout);
 #endif
 
-
   // clone the input image so that we have a copy of the original
   Mat img_cpy = img.clone();
 
@@ -632,8 +644,8 @@ int_t opencv_dot_targets(Mat & img,
   //  cv::THRESH_TRIANGLE = 16
   int threshold_mode = options.get<int_t>(opencv_server_threshold_mode,0);
   //std::cout << "opencv_dot_targets(): option, threshold mode:   " << threshold_mode << std::endl;
-  int min_blob_size = 100;
-
+  int min_blob_size = options.get<int_t>(opencv_server_min_blob_size,100);
+  DEBUG_MSG("using min_blob_size: " << min_blob_size);
   // establish the calibration plate properties
   TEUCHOS_TEST_FOR_EXCEPTION(!options.isParameter(DICe::num_cal_fiducials_x),std::runtime_error,"");
   TEUCHOS_TEST_FOR_EXCEPTION(!options.isParameter(DICe::num_cal_fiducials_y),std::runtime_error,"");
@@ -698,17 +710,19 @@ int_t opencv_dot_targets(Mat & img,
   // chances are that this indicates p thresholding problem to begin with
   if (key_points.size() != 3) {
     // try again to see if the markers a too small, if so enable smaller blob sizes and try again
-    min_blob_size = 10;
-    get_dot_markers(img_cpy,key_points,i_thresh,invert,options,min_blob_size);
-    if(key_points.size() !=3){
-      // try a larger min blob size
-      min_blob_size = 500;
-      get_dot_markers(img_cpy,key_points,i_thresh,invert,options,min_blob_size);
-      if(key_points.size() != 3){
-        std::cout << "*** warning: unable to identify three keypoints, other points will not be extracted" << std::endl;
-        keypoints_found = false;
-      }
-    }
+//    min_blob_size = 10;
+//    get_dot_markers(img_cpy,key_points,i_thresh,invert,options,min_blob_size);
+//    if(key_points.size() !=3){
+      std::cout << "*** warning: unable to identify three keypoints, other points will not be extracted" << std::endl;
+      keypoints_found = false;
+//      // try a larger min blob size
+//      min_blob_size = 500;
+//      get_dot_markers(img_cpy,key_points,i_thresh,invert,options,min_blob_size);
+//      if(key_points.size() != 3){
+//        std::cout << "*** warning: unable to identify three keypoints, other points will not be extracted" << std::endl;
+//        keypoints_found = false;
+//      }
+//    }
   }
   Point cvpoint;
   if(preview_thresh){
@@ -856,6 +870,7 @@ void get_dot_markers(cv::Mat img,
 
   // setup the blob detector
   SimpleBlobDetector::Params params;
+  params.filterByArea = true;
   params.maxArea = 10e4;
   params.minArea = min_size;
   cv::Ptr<cv::SimpleBlobDetector> detector = cv::SimpleBlobDetector::create(params);
@@ -884,7 +899,32 @@ void get_dot_markers(cv::Mat img,
   // detect dots on the appropriately inverted image
   if (invert) detector->detect(not_src, keypoints);
   else detector->detect(bi_src, keypoints);
+  DEBUG_MSG("get_dot_markers(): preliminary num keypoints " << keypoints.size());
+  if(keypoints.size()==0) return;
+  // the diameters of the keypoints should be within 30% of each other
+  float avg_diameter = 0.0f;
+  for(size_t i=0;i<keypoints.size();++i){
+    avg_diameter += keypoints[i].size;
+  }
+  avg_diameter /= keypoints.size();
+  DEBUG_MSG("get_dot_markers(): avg keypoint diameter " << avg_diameter);
+
+  size_t i = keypoints.size();
+  while (i--) {
+//    DEBUG_MSG("get_dot_markers(): possible keypoint " << i << " diameter " << keypoints[i].size);
+    // remove the keypoint from the vector
+    if(keypoints[i].size<=0.0){
+//      DEBUG_MSG("get_dot_markers(): removing keypoint " << i << " due to keypoint size == 0.0");
+      keypoints.erase(keypoints.begin() + i);
+    }
+    if(std::abs(keypoints[i].size-avg_diameter)/avg_diameter>0.30){
+//      DEBUG_MSG("get_dot_markers(): removing keypoint " << i << " due to keypoint size >30% difference in diameter from avg");
+      keypoints.erase(keypoints.begin() + i);
+    }
+  }
   DEBUG_MSG("get_dot_markers(): num keypoints " << keypoints.size());
+//  for (size_t i = 0;i<keypoints.size();++i)
+//    DEBUG_MSG("get_dot_markers(): keypoint " << i << " diameter " << keypoints[i].size);
 }
 
 //calculate the transformation coefficients

src/opencv/DICe_OpenCVServerUtils.h

@@ -78,6 +78,7 @@ const char* const opencv_server_threshold_start = "threshold_start";
 const char* const opencv_server_threshold_end = "threshold_end";
 const char* const opencv_server_threshold_step = "threshold_step";
 const char* const opencv_server_block_size = "block_size";
+const char* const opencv_server_min_blob_size = "min_blob_size";
 const char* const opencv_server_binary_constant = "binary_constant";
 const char* const opencv_server_dot_tol = "dot_tol";
 const char* const opencv_server_use_adaptive_threshold = "use_adaptive_threshold";