From 36eb9db9d80f8d48496f9bc872b85d9eccd3d69a Mon Sep 17 00:00:00 2001
From: Reko Vuorinen <ravuor@utu.fi>
Date: Thu, 24 Apr 2025 14:59:04 +0300
Subject: [PATCH] Tried to fix the ARTag detection
---
.../tello_racing_pkg/tello_drone_race.py | 209 +++++++-----------
1 file changed, 81 insertions(+), 128 deletions(-)
diff --git a/src/drone_racing_ros2/tello_ros/tello_racing_pkg/tello_racing_pkg/tello_drone_race.py b/src/drone_racing_ros2/tello_ros/tello_racing_pkg/tello_racing_pkg/tello_drone_race.py
index bd3e581..7e638c3 100755
--- a/src/drone_racing_ros2/tello_ros/tello_racing_pkg/tello_racing_pkg/tello_drone_race.py
+++ b/src/drone_racing_ros2/tello_ros/tello_racing_pkg/tello_racing_pkg/tello_drone_race.py
@@ -92,7 +92,7 @@ class GateNavigator(Node):
self.get_logger().info(">>> Running in FLIGHT MODE <<<")
self.get_logger().info(f"Subscribed to {image_topic}, publishing to {cmd_vel_topic}.")
self.get_logger().info(f"Green HSV range: {self.hsv_lower_green} to {self.hsv_upper_green}")
- self.get_logger().info(f"Wood HSV range: {self.hsv_lower_wood} to {self.hsv_upper_wood}")
+
# Initial Takeoff
@@ -177,41 +177,82 @@ class GateNavigator(Node):
if self.check_and_land_on_red(red_mask):
return # Land triggered, skip rest
- contours, _ = cv2.findContours(green_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
-
- self.gate_detected = False
+ ar_tag_active = False
+ gray = cv2.cvtColor(cv_image, cv2.COLOR_BGR2GRAY)
+ aruco_dict = aruco.Dictionary_get(aruco.DICT_4X4_50)
+ parameters = aruco.DetectorParameters_create()
+ # (same fine-tuning you already had)
+ parameters.adaptiveThreshWinSizeMin = 5
+ parameters.adaptiveThreshWinSizeMax = 25
+ parameters.adaptiveThreshWinSizeStep = 5
+ parameters.adaptiveThreshConstant = 7
+ parameters.minMarkerPerimeterRate = 0.02
+ parameters.maxMarkerPerimeterRate = 4.0
+ parameters.polygonalApproxAccuracyRate = 0.04
+ parameters.minCornerDistanceRate = 0.05
+ parameters.minMarkerDistanceRate = 0.05
+ parameters.cornerRefinementMethod = aruco.CORNER_REFINE_SUBPIX
+ parameters.errorCorrectionRate = 0.7
+
+ corners, ids, _ = aruco.detectMarkers(gray, aruco_dict,
+ parameters=parameters)
+
+ self.gate_detected = False
best_cx, best_cy = self.image_center_x, self.image_center_y
- # -------------------------------------------------------------------------
- # NEW LOGIC – handle 3 + separate green blobs by averaging their centers
- # -------------------------------------------------------------------------
+ if ids is not None:
+ ar_tag_active = True
+ tag_centers = []
+ detected_ids_list = ids.flatten().tolist()
+ for i, corner_set in enumerate(corners):
+ c = corner_set[0] # shape (4,2)
+ cx, cy = int(np.mean(c[:, 0])), int(np.mean(c[:, 1]))
+ tag_centers.append((cx, cy))
+ cv2.circle(cv_image, (cx, cy), 5, (0, 255, 255), -1)
+ cv2.putText(cv_image, str(detected_ids_list[i]),
+ (cx + 10, cy + 10), cv2.FONT_HERSHEY_SIMPLEX,
+ 0.5, (255, 0, 255), 2)
+
+ best_cx = int(np.mean([p[0] for p in tag_centers]))
+ best_cy = int(np.mean([p[1] for p in tag_centers]))
+ cv2.circle(cv_image, (best_cx, best_cy), 7, (255, 0, 255), -1)
+ aruco.drawDetectedMarkers(cv_image, corners, ids)
+
+ self.gate_detected = True
+ self.passing_gate = True
+ self.last_gate_time = self.get_clock().now()
+
+ if not ar_tag_active:
+ contours, _ = cv2.findContours(green_mask, cv2.RETR_EXTERNAL,
+ cv2.CHAIN_APPROX_SIMPLE)
+
+
if contours:
# only keep contours whose area is significant for a gate candidate
- sig_contours = [c for c in contours if cv2.contourArea(c) > self.min_contour_area] # << mod
-
- if len(sig_contours) >= 3: # << mod
- centers = [] # << mod
- for c in sig_contours: # << mod
- M = cv2.moments(c) # << mod
- if M["m00"] == 0: # << mod
- continue # << mod
- cx = int(M["m10"] / M["m00"]) # << mod
- cy = int(M["m01"] / M["m00"]) # << mod
- centers.append((cx, cy)) # << mod
- cv2.drawContours(cv_image, [c], -1, (0, 255, 0), 2) # << mod
-
- if centers: # << mod
- best_cx = int(np.mean([p[0] for p in centers])) # << mod
- best_cy = int(np.mean([p[1] for p in centers])) # << mod
- self.gate_detected = True # << mod
- self.passing_gate = True # << mod
- self.last_gate_time = self.get_clock().now() # << mod
- cv2.circle(cv_image, (best_cx, best_cy), 7, (0, 0, 255), -1) # << mod
- cv2.putText(cv_image, f"Gate ({best_cx},{best_cy})", # << mod
- (best_cx - 40, best_cy - 20), cv2.FONT_HERSHEY_SIMPLEX, # << mod
- 0.5, (255, 255, 255), 2) # << mod
- else: # << mod
- # ---------- original single-contour logic (unchanged) ----------
+ sig_contours = [c for c in contours if cv2.contourArea(c) > self.min_contour_area]
+
+ if len(sig_contours) >= 3:
+ centers = []
+ for c in sig_contours:
+ M = cv2.moments(c)
+ if M["m00"] == 0:
+ continue
+ cx = int(M["m10"] / M["m00"])
+ cy = int(M["m01"] / M["m00"])
+ centers.append((cx, cy))
+ cv2.drawContours(cv_image, [c], -1, (0, 255, 0), 2)
+
+ if centers:
+ best_cx = int(np.mean([p[0] for p in centers]))
+ best_cy = int(np.mean([p[1] for p in centers]))
+ self.gate_detected = True
+ self.passing_gate = True
+ self.last_gate_time = self.get_clock().now()
+ cv2.circle(cv_image, (best_cx, best_cy), 7, (0, 0, 255), -1)
+ cv2.putText(cv_image, f"Gate ({best_cx},{best_cy})",
+ (best_cx - 40, best_cy - 20), cv2.FONT_HERSHEY_SIMPLEX,
+ 0.5, (255, 255, 255), 2)
+ else:
largest_contour = max(sig_contours, key=cv2.contourArea) if sig_contours else None
if largest_contour is not None:
M = cv2.moments(largest_contour)
@@ -228,9 +269,7 @@ class GateNavigator(Node):
cv2.circle(cv_image, (cx, cy), 7, (0, 0, 255), -1)
cv2.putText(cv_image, f"Gate ({cx},{cy})", (cx - 40, cy - 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 255), 2)
- # -------------------- end of new / modified section ----------------------
- # ---------------- remainder of function is UNCHANGED --------------------
# Control Logic Calculation
twist_msg = Twist()
@@ -266,100 +305,14 @@ class GateNavigator(Node):
self.passing_gate = False
else:
- # ARTag detection
- gray = cv2.cvtColor(cv_image, cv2.COLOR_BGR2GRAY)
- aruco_dict = aruco.Dictionary_get(aruco.DICT_4X4_50) # VERIFY THIS!
-
- # Detector Parameters
- parameters = aruco.DetectorParameters_create()
-
- #--------------------------------------------------------------------------------------------
- # Thresholding Parameters
- # Lowering min window size can help detect smaller tags
- parameters.adaptiveThreshWinSizeMin = 5 # Default: 3
- # Increasing max window size can help with varying lighting
- parameters.adaptiveThreshWinSizeMax = 25 # Default: 23
- # Step size for adaptive thresholding windows
- parameters.adaptiveThreshWinSizeStep = 5 # Default: 10
- # Constant subtracted from mean during adaptive thresholding
- parameters.adaptiveThreshConstant = 7 # Default: 7
-
- # Contour Filtering Parameters
- # Minimum perimeter relative to image dimensions (decrease if tags are very small)
- parameters.minMarkerPerimeterRate = 0.02 # Default: 0.03 (Value depends heavily on expected tag size in image)
- # Maximum perimeter relative to image dimensions (increase if tags can be very large/close)
- parameters.maxMarkerPerimeterRate = 4.0 # Default: 4.0
- # Accuracy for polygonal approximation (small adjustments might help)
- parameters.polygonalApproxAccuracyRate = 0.04 # Default: 0.03
- # Minimum distance between corners (increase slightly if getting false positives from noise)
- parameters.minCornerDistanceRate = 0.05 # Default: 0.05
- # Minimum distance between different markers (increase if markers are close and misidentified)
- parameters.minMarkerDistanceRate = 0.05 # Default: 0.05
-
- # Corner Refinement
- # Use sub-pixel refinement for potentially better accuracy (but slightly slower)
- parameters.cornerRefinementMethod = aruco.CORNER_REFINE_SUBPIX # Default: CORNER_REFINE_NONE
- # Other options: aruco.CORNER_REFINE_CONTOUR, aruco.CORNER_REFINE_APRILTAG (if using AprilTags)
-
- # Error Correction
- # Allow slightly more error correction (might help with damaged/occluded tags, but increases false positive risk)
- parameters.errorCorrectionRate = 0.7 # Default: 0.6
- #-------------------------------------------------------------------------------------------------------------
-
- corners, ids, rejected = aruco.detectMarkers(gray, aruco_dict, parameters=parameters) # Rejected candidates for debugging
-
- # Handle 1 or more tags
- if ids is not None:
- tag_centers = []
- detected_ids_list = ids.flatten().tolist() # Get a flat list of detected IDs
-
- for i, corner_set in enumerate(corners):
- c = corner_set[0] # shape (4,2)
- cx = int(np.mean(c[:, 0]))
- cy = int(np.mean(c[:, 1]))
- tag_centers.append((cx, cy))
- # Draw tag center and ID
- cv2.circle(cv_image, (cx, cy), 5, (0, 255, 255), -1)
- cv2.putText(cv_image, str(detected_ids_list[i]), (cx + 10, cy + 10),
- cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 0, 255), 2)
-
-
- # Calculate the average center of ALL detected tags
- avg_cx = int(np.mean([p[0] for p in tag_centers]))
- avg_cy = int(np.mean([p[1] for p in tag_centers]))
-
- error_x = avg_cx - self.image_center_x
- error_y = avg_cy - self.image_center_y
-
- calculated_angular_z = self.kp_yaw * float(error_x)
- calculated_linear_z = self.kp_alt * float(error_y)
- # Maintain forward speed when tags are detected
- calculated_linear_x = self.target_forward_speed
-
- twist_msg.angular.z = calculated_angular_z
- twist_msg.linear.z = calculated_linear_z
- twist_msg.linear.x = calculated_linear_x
-
- num_tags = len(ids)
- self.get_logger().info(f"{num_tags} AR Tag(s) Found (IDs: {detected_ids_list})! Center:({avg_cx},{avg_cy}), Err:({error_x},{error_y}), Cmd: lin.x={twist_msg.linear.x:.2f}, lin.z={twist_msg.linear.z:.2f}, ang.z={twist_msg.angular.z:.2f}", throttle_duration_sec=1.0)
-
- # Set passing_gate mode active (even if only one tag is seen)
- self.passing_gate = True
- self.last_gate_time = self.get_clock().now()
-
- # Visualize overall target center and detected markers
- cv2.circle(cv_image, (avg_cx, avg_cy), 7, (255, 0, 255), -1) # Magenta for overall center
- aruco.drawDetectedMarkers(cv_image, corners, ids) # Draw borders
-
- else:
- # No Gate (Color) or AR Tags Found -> Search Rotation
- search_twist = Twist()
- search_twist.angular.z = 0.3
- search_twist.linear.x = 0.05 # Optional small forward movement during search
- self.get_logger().info("No gate (Color/AR Tag) detected — Search Rotation.", throttle_duration_sec=1.0)
- twist_msg = search_twist
- # Ensure passing_gate is false if we enter search mode
- # self.passing_gate = False # This might interfere with the timeout logic, review if needed
+ # No Gate (Color) or AR Tags Found -> Search Rotation
+ search_twist = Twist()
+ search_twist.angular.z = 0.3
+ search_twist.linear.x = 0.05 # Optional small forward movement during search
+ self.get_logger().info("No gate (Color/AR Tag) detected — Search Rotation.", throttle_duration_sec=1.0)
+ twist_msg = search_twist
+ # Ensure passing_gate is false if we enter search mode
+ # self.passing_gate = False # This might interfere with the timeout logic, review if needed
# Publish Command (Conditional)
--
GitLab