Bug fix and tuning (#62)

* Draw boundary on top of all detections

* Update launch

* Add bag name to video

* Update params for better evaluation

* Fix OBB bug

* Update params

* Tune transform

Author: adrian-soch

Metrics/preprocessGT.py

@@ -35,16 +35,14 @@ def main(args):
     write_lines(result_name, normalized_lines)
 
     # Define a list of points as tuples, must be clockwise
-    # q7_2_may10_2023_clean bag
-    # roi_points = np.array([(495, 230), (900, 230), (1910, 730), (525, 900)])
 
-    # q7_2_may10_2023_clean_short_range
+    # may10 short_range
     # roi_points = np.array([(658,305), (980,960), (1900,737), (1049,287)])
 
-    # oct18_r9
+    # oct18
     roi_points = np.array([(5,735),(1270,645),(1500,676),(240,1060)])
 
-    # dec7_dhd1_short
+    # dec7
     # roi_points = np.array([(835,480), (1050,1056), (1918,845), (1130,465)])
 
     # Create a polygon object from the points using shapely
@@ -56,7 +54,6 @@ def main(args):
     if args.image_path is not None:
         # Draw roi on image
         img = cv2.imread(args.image_path)
-        cv2.polylines(img, [roi_points], True, (255, 0, 0), 3)
 
     # Read the csv file using pandas
     df = pd.read_csv(result_name, header=None)
@@ -85,6 +82,10 @@ def main(args):
         if args.image_path is not None:
             cv2.polylines(img, [points], True, colour, 2)
 
+    if args.image_path is not None:
+        # Draw roi on image
+        cv2.polylines(img, [roi_points], True, (255, 0, 0), 3)
+
     # Drop the rows that are in the remove list using pandas
     df = df.drop(remove_indices)
 

fusion_module/fusion_module/fusion_visualizer_node.py

@@ -23,6 +23,12 @@ def __init__(self):
         self.declare_parameter('save_video', False)
         save_video = self.get_parameter(
             'save_video').get_parameter_value().bool_value
+        
+        self.declare_parameter('video_name', 'video')
+        self.video_name = self.get_parameter(
+            'video_name').get_parameter_value().string_value
+        
+        print(self.video_name + '\n')
 
         self.bridge = CvBridge()
         self.image_sub = Subscriber(self, Image, 'image')
@@ -104,7 +110,7 @@ def callback(self, image_msg, detection_msg):
         if self.video_writer is None:
             fourcc = VideoWriter_fourcc(*'mp4v')  # choose codec
             fps = 10  # choose frame rate
-            filename = 'video_{}.mp4'.format(datetime.datetime.now().strftime(
+            filename = self.video_name + '_{}.mp4'.format(datetime.datetime.now().strftime(
                 '%Y%m%d_%H%M%S'))  # choose file name with current timestamp
             self.video_writer = VideoWriter(
                 filename, fourcc, fps, (cv_image.shape[1], cv_image.shape[0]))

lidar_pipeline/configs/dec7_config.yaml

@@ -40,7 +40,7 @@ projection_node:
     # Transform from Lidar to camera frame
     lidar2cam_extrinsic:
       rotation: [0.0, 1.0, 0.0, 0.0, 0.0, -1.0, -1.0, 0.0, 0.0]
-      translation: [0.00, 0.0508, 0.0502]
+      translation: [-0.3, -0.32, 0.0502]
 
     # Transform from lidar data from to lidar sensor frame
     # See https://github.com/ros-drivers/ros2_ouster_drivers/issues/87

lidar_pipeline/configs/oct18_r1_config.yaml

@@ -0,0 +1,55 @@
+# These parameters are dependant on the sensor location
+# Note: For converting angles use https://www.andre-gaschler.com/rotationconverter/
+perception_node:
+  ros__parameters:
+
+    # Transformation from sensor to world frame
+    lidar2world_transform:
+      # [x,y,z] in meters
+      translation: [0.0, 0.0, 3.25]
+
+      # [w,x,y,z] normalized quaternion
+      quaternion: [0.999769, 0.0, -0.021498, 0.0]
+
+    # The crop box isolates a region of interest in the pointcloud
+    # all point outside this box are removed
+    # Use the roi_visulaizer.py tool to find a new crop_box params if needed
+    crop_box_transform:
+      # [x,y,z] in meters
+      translation: [35.0, -0.5, 1.75]
+
+      # [w,x,y,z] normalized quaternion
+      quaternion: [0.9025852843498606, 0.0, 0.0, 0.43051109680829514]
+
+      # [x, y, z] in meters
+      size:  [16.0, 75.0, 3.5]
+
+
+# Use the same `lidar2world_transform` for both nodes
+projection_node:
+  ros__parameters:
+
+    # Matrix from intrinsic camera calibration
+    # This day's data was recorded at 1920x1080
+    camera_matrix: [1199.821557, 0.0, 960.562236, 0.0, 1198.033465, 551.675808, 0.0, 0.0, 1.0]
+
+    # Transformation from sensor to world frame
+    lidar2world_transform:
+      # [x,y,z] in meters
+      translation: [0.0, 0.0, 3.25]
+
+      # [w,x,y,z] normalized quaternion
+      quaternion: [0.999769, 0.0, -0.021498, 0.0]
+
+    # Transform from Lidar to camera frame
+    lidar2cam_extrinsic:
+      rotation: [-0.0174524,  0.9998477,  0.0000000,-0.0000000, -0.0000000, -1.0000000,-0.9998477, -0.0174524,  0.0000000 ]
+      translation: [0.00, 0.0508, 0.0502]
+
+    # Transform from lidar data from to lidar sensor frame
+    # See https://github.com/ros-drivers/ros2_ouster_drivers/issues/87
+    # The translation part may already be accounted for internally in the driver
+    lidarData2lidarSensor_extrinsic:
+      rotation: [-1.0, 0.0, 0.0, 0.0, -1.0, 0.0, 0.0, 0.0, 1.0]
+      translation: [0.0, 0.0, 0.03618]
+

lidar_pipeline/launch/fusion_demo.launch.py

@@ -19,7 +19,7 @@
 # Used to change playback rate of ros bag
 # A9 data bags were recoreded at 2.5Hz so they need a x4 speedup
 # if left as 1 then thre is no rate change
-BAG_PLAY_RATE = 0.16
+BAG_PLAY_RATE = 1.0
 FLIP_IMAGE = False
 
 BAG_PLAY_LOOP = True
@@ -35,16 +35,16 @@
 '''
 ABS_PATH_TO_ROSBAGS = '/home/adrian/dev/bags/'
 
-# 10, 7, 6, 12
-BAG_SELECTOR = 7
+# 10, 7, 6, 12, 13
+BAG_SELECTOR = 13
 
-# Determines what kind of output you want, Video/Rviz2
+# Determines what kind of output you want, Video/Rviz2/csv_tracker_data
 SAVE_OUTPUT_VIDEO = True
 SAVE_CSV_FUSION_OUTPUT = True
 SHOW_RVIZ = False
 
 # Fusion Overrides
-LIDAR_RESULT_ONLY = True
+LIDAR_RESULT_ONLY = False
 CAM_RESULT_ONLY = False
 
 # Because of the yolov5 includes, its easier to just run this directly
@@ -58,7 +58,6 @@
     # FLIP_IMAGE = True
     CONFIG_NAME = 'default_config.yaml'
 
-
 # MARC Rooftop data without data syncronization
 # Fusion and projection will not work
 elif BAG_SELECTOR == 0:
@@ -77,6 +76,9 @@
 elif BAG_SELECTOR == 12:
     BAG_NAME = 'cleaned_bags/may10_r5_clean'
     CONFIG_NAME = 'may10_config.yaml'
+elif BAG_SELECTOR == 13:
+    BAG_NAME = 'cleaned_bags/oct18_r1_clean'
+    CONFIG_NAME = 'oct18_r1_config.yaml'
 
 # MARC Rooftop data with syncronized lidar + camera
 elif BAG_SELECTOR == 2:
@@ -122,6 +124,15 @@
     print('Invalid bag selection!')
     exit(-1)
 
+# Set and validate parameters
+
+if SAVE_CSV_FUSION_OUTPUT:
+    BAG_PLAY_RATE = 0.16
+
+if LIDAR_RESULT_ONLY and CAM_RESULT_ONLY:
+    print('Error: Must use lidar or camera results for fusion.')
+    exit(ValueError)
+
 START_BAG_DELAY = 0.0
 if SAVE_OUTPUT_VIDEO:
     START_BAG_DELAY = 7.5
@@ -196,13 +207,15 @@ def generate_launch_description():
         ]
     )
 
+    video_name = BAG_NAME.split('/')[-1]
     fusion_viz = Node(
         package='fusion_module',
         executable='fusion_viz_node',
         name='fusion_viz_node',
         parameters=[
             {'flip_image': FLIP_IMAGE},
             {'save_video': SAVE_OUTPUT_VIDEO},
+            {'video_name': video_name},
         ]
     )
 

lidar_pipeline/src/lidar_processing.cpp

@@ -277,9 +277,9 @@ vision_msgs::msg::BoundingBox3D LidarProcessing::getOrientedBoudingBox(const pcl
     size.getArray3fMap() = max_pt_T.getArray3fMap() - min_pt_T.getArray3fMap();
 
     vision_msgs::msg::BoundingBox3D bbox;
-    bbox.center.position.x    = centroid[X];
-    bbox.center.position.y    = centroid[Y];
-    bbox.center.position.z    = centroid[Z];
+    bbox.center.position.x    = center[X];
+    bbox.center.position.y    = center[Y];
+    bbox.center.position.z    = center[Z];
     bbox.center.orientation.x = bboxQ.x();
     bbox.center.orientation.y = bboxQ.y();
     bbox.center.orientation.z = bboxQ.z();