Increased number of hard detections

src/feature_matching.py

@@ -18,7 +18,7 @@ def find_matches(des1, des2):
     # Keep only good matches
     good_matches = []
     for match1, match2 in matches:
-        if match1.distance < 0.7 * match2.distance:
+        if match1.distance < 0.8 * match2.distance:
             good_matches.append(match1)
 
     return good_matches

src/hard_pipeline/feature_matching.py

@@ -18,7 +18,7 @@ def find_matches(des1, des2):
     # Keep only good matches
     good_matches = []
     for match1, match2 in matches:
-        if match1.distance < 0.8 * match2.distance:
+        if match1.distance < 0.7 * match2.distance:
             good_matches.append(match1)
 
     return good_matches

src/hard_pipeline/main.py

@@ -28,8 +28,8 @@ def preprocess_box_hard(b):
     pr_box = b.copy()
     pr_box = image_processing.convert_grayscale(pr_box)
     pr_box = image_processing.equalize_histogram(pr_box)
-    #if pr_box.shape[1] >= 200:
-        #pr_box = image_processing.resize_img(pr_box, 0.5)
+    if pr_box.shape[1] >= 200:
+        pr_box = image_processing.blur_image(pr_box)
 
     return pr_box
 
@@ -43,7 +43,7 @@ def preprocess_box_hard(b):
 
 for scene in scenes:
     visualization_scene = scene.copy()
-    visualization_scene = image_processing.resize_img(visualization_scene, 2)
+    visualization_scene = image_processing.resize_img(visualization_scene, 3)
 
     sub_scenes = parallel_hough.split_shelves(scene)
 

src/hard_pipeline/object_validation.py

@@ -24,7 +24,7 @@ def compute_rectangularity(bounds):
 
 
 def check_rectangularity(bounds):
-    return compute_rectangularity(bounds) > 0.3
+    return compute_rectangularity(bounds) > 0.7
 
 
 def is_contained(bounds_outer, bounds_inner):
@@ -129,10 +129,10 @@ def validate_color(box, scene, used_box_pts, used_scene_pts, match_bounds, homog
     t = box_masked_area[box_masked_area > 0]
     area_ratio = t.shape[0] / (masked_box.shape[0] * masked_box.shape[1])
 
-    if area_ratio < 0.20:
+    if area_ratio < 0.05:
         return False
 
-    return compare_hue(masked_box, masked_scene, homography, match_bounds)
+    return compare_hue(box_val, scene_val, homography, match_bounds)
 
 
 # Compares the colors of the template with the instance found in the scene, to test the compatibility of the match. This
@@ -142,6 +142,8 @@ def validate_color(box, scene, used_box_pts, used_scene_pts, match_bounds, homog
 # discarded.
 def compare_hue(box, scene, homography, match_bounds):
     transformed_box, test_scene = image_processing.transform_box_in_scene(box, scene, homography)
+    #visualization.display_img(transformed_box)
+    #visualization.display_img(test_scene)
 
     left = max(0, np.min([match_bounds[0][0], match_bounds[1][0], match_bounds[2][0], match_bounds[3][0]]))
     right = min(transformed_box.shape[1], np.max([match_bounds[0][0], match_bounds[1][0], match_bounds[2][0], match_bounds[3][0]]))
@@ -209,4 +211,4 @@ def compare_hue(box, scene, homography, match_bounds):
         if np.isin(peak, peaks1):
             common_peaks = common_peaks + 1
 
-    return common_peaks >= 2 and hue_comparison > 0.8
+    return common_peaks >= 2 and hue_comparison > 0.6

src/hard_pipeline/parallel_hough.py

@@ -10,9 +10,9 @@
 
 import feature_detection
 import image_processing
-import object_validation
 import prove
 import visualization
+from hard_pipeline import object_validation
 
 low_H = 90
 low_S = 8
@@ -91,7 +91,6 @@ def preprocess_sub_scene(s):
     pr_scene = s.copy()
     pr_scene = image_processing.convert_grayscale(pr_scene)
     pr_scene = image_processing.equalize_histogram(pr_scene)
-    pr_scene = image_processing.sharpen_img(pr_scene)
 
     return pr_scene
 
@@ -112,9 +111,9 @@ def erase_bar(matches_mask, bounds):
 
 def compute_sub_image(dict_box_features, sub_image):
     sub_scene, y = sub_image
-    sub_scene = image_processing.resize_img(sub_scene, 2)
+    sub_scene = image_processing.resize_img(sub_scene, 3)
     proc_scene = preprocess_sub_scene(sub_scene)
-    y *= 2
+    y *= 3
     test_scene = image_processing.resize_img_dim(sub_scene, proc_scene.shape[1], proc_scene.shape[0])
 
     s = time.time()
@@ -171,7 +170,7 @@ def compute_sub_image(dict_box_features, sub_image):
                         bounds_dict[gray_index] = bar_intersecting
                     if result == -1:  # c'è intersezione ma i box non sono completamente uno dentro l'altro, effettuo ulteriore controllo con compare_detections_hard
                         #print('Calling compare_detections_hard')
-                        result = object_validation.compare_detections_hard(bar_intersected, bar_intersecting, 1.5)
+                        result = object_validation.compare_detections_hard(bar_intersected, bar_intersecting, 2)
                         #print('Result of compare_detections_hard = ', result)
                         # if result == 1 allora c'è troppa sovrapposizione tra i box e l'intersecato è migliore dell'intersecante, non faccio nulla
                         if result == 0:  # c'è troppa sovrapposizione tra i box e l'intersecante è migliore dell'intersecato, sostituisco