Add shape matching to cam_vision

grp_astro/CMakeLists.txt

@@ -37,6 +37,12 @@ install(DIRECTORY
   DESTINATION share/${PROJECT_NAME}/
 )
 
+# Copy images
+install(DIRECTORY
+  src/img
+  DESTINATION lib/${PROJECT_NAME}
+)
+
 # Python scripts
 install( PROGRAMS 
   src/cam_driver

grp_astro/src/cam_vision

@@ -4,7 +4,7 @@ from rclpy.node import Node
 from sensor_msgs.msg import Image
 from std_msgs.msg import String
 import  numpy as np
-import sys, cv2
+import sys, cv2, os, pathlib
 from cv_bridge import CvBridge
 
 
@@ -14,9 +14,9 @@ class CameraVision(Node):
         super().__init__(name) # Create the node
 
         # Initialiaze node variables
-        self._low = np.array([58-20, 60, 40])
-        self._high = np.array([58+20, 255,255])
-        self._kernel = np.ones((3, 3), np.uint8)
+        self._low = np.array([72-25, 107-55, 84-55])
+        self._high = np.array([65+25, 219+30, 255])
+        self._kernel = np.ones((5, 5), np.uint8)
 
         self._color_info = (0, 0, 255)
 
@@ -42,6 +42,28 @@ class CameraVision(Node):
         self._rgb_img = self._bridge.imgmsg_to_cv2(img, desired_encoding='passthrough')
 
 
+    def publish_msgs(self, image, mask):
+        # Convert cv2 images to ROS2 RGB Image20
+        ros2_mask_image = cv2.cvtColor(mask, cv2.COLOR_GRAY2RGB)
+        ros2_mask_image = cv2.cvtColor(ros2_mask_image, cv2.COLOR_RGB2BGR)
+        ros2_mask_image = self._bridge.cv2_to_imgmsg(ros2_mask_image,"bgr8")
+        ros2_mask_image.header.stamp = self.get_clock().now().to_msg()
+        ros2_mask_image.header.frame_id = "image"
+
+        ros2_rgb_image = cv2.cvtColor(image, cv2.COLOR_HSV2BGR)
+        ros2_rgb_image = self._bridge.cv2_to_imgmsg(ros2_rgb_image,"bgr8")
+        ros2_rgb_image.header.stamp = self.get_clock().now().to_msg()
+        ros2_rgb_image.header.frame_id = "image"
+
+
+        # Publish String message -> to inform that we have detected a bottle
+        str = String()
+        str.data = "A bottle has been detected"
+        self._detection_publisher.publish(str)
+        # Publish Images -> to be able to detect false positives and improve the algorithm
+        self._rgb_detection_publisher.publish(ros2_rgb_image)
+        self._mask_detection_publisher.publish(ros2_mask_image)
+        
     # Node internal loop
     def loop(self):
         if self._rgb_img is None :
@@ -58,37 +80,45 @@ class CameraVision(Node):
         mask = cv2.dilate(mask, self._kernel, iterations=4)
 
 
-        # Segmentation using the "Min enclosing circle" method
-        elements = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)[-2]
-        if len(elements) > 0: # If we detect elements
-            c = max(elements, key=cv2.contourArea)
+        # Segmentation using shape matching
+        directory = pathlib.PurePath(__file__).parent
+        pattern = cv2.imread(str(directory / 'img' / 'template.jpg'), cv2.IMREAD_GRAYSCALE)
+        pattern2 = cv2.imread(str(directory / 'img' / 'template2.jpg'), cv2.IMREAD_GRAYSCALE)
+
+        contours,_ = cv2.findContours(pattern, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
+        sorted_contours = sorted(contours, key=cv2.contourArea, reverse=True)
+        template_contour = sorted_contours[0]
+
+        contours,_ = cv2.findContours(pattern2, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
+        sorted_contours = sorted(contours, key=cv2.contourArea, reverse=True)
+        template2_contour = sorted_contours[0]
+        
+        contours,_ = cv2.findContours(mask, cv2.RETR_CCOMP, cv2.CHAIN_APPROX_SIMPLE)
+        sorted_contours = sorted(contours, key=cv2.contourArea, reverse=True)
+        
+
+        for c in contours:
+        
+            # Compare our mask with our templates
+            match = cv2.matchShapes(template_contour, c, 3, 0.0)
+            match2 = cv2.matchShapes(template2_contour, c, 3, 0.0)
+
+            # Get size of object
             (x, y), radius = cv2.minEnclosingCircle(c)
             
-            if radius > 20: #If the radius of the biggest element detected is greater than 30, we consider that we have detected a bottle
+            # If we find a correct match that is big enough, we publish data
+            if min(match, match2) < 0.2 and radius > 30:
+                print(f'Match : {min(match, match2)}')
+
                 # Circle the bottle on the image 
                 cv2.circle(image, (int(x), int(y)), int(radius), self._color_info, 2)
                 cv2.circle(image, (int(x), int(y)), 5, self._color_info, 10)
 
-                # Convert cv2 images to ROS2 RGB Image
-                ros2_mask_image = cv2.cvtColor(mask, cv2.COLOR_GRAY2RGB)
-                ros2_mask_image = cv2.cvtColor(ros2_mask_image, cv2.COLOR_RGB2BGR)
-                ros2_mask_image = self._bridge.cv2_to_imgmsg(ros2_mask_image,"bgr8")
-                ros2_mask_image.header.stamp = self.get_clock().now().to_msg()
-                ros2_mask_image.header.frame_id = "image"
-
-                ros2_rgb_image = cv2.cvtColor(image, cv2.COLOR_HSV2BGR)
-                ros2_rgb_image = self._bridge.cv2_to_imgmsg(ros2_rgb_image,"bgr8")
-                ros2_rgb_image.header.stamp = self.get_clock().now().to_msg()
-                ros2_rgb_image.header.frame_id = "image"
+                # Publish images and string message
+                self.publish_msgs(image, mask)
 
+                break # Stop checking other contours
 
-                # Publish String message -> to inform that we have detected a bottle
-                str = String()
-                str.data = "A bottle has been detected"
-                self._detection_publisher.publish(str)
-                # Publish Images -> to be able to detect false positives and improve the algorithm
-                self._rgb_detection_publisher.publish(ros2_rgb_image)
-                self._mask_detection_publisher.publish(ros2_mask_image)