def callback_with_cluster_box(self, cluster_boxes_msg, instance_boxes_msg, instance_label_msg):
labeled_cluster_boxes = BoundingBoxArray()
labeled_instance_boxes = BoundingBoxArray()
labeled_cluster_boxes.header = cluster_boxes_msg.header
labeled_instance_boxes.header = instance_boxes_msg.header
for index, box in enumerate(cluster_boxes_msg.boxes):
if not box.pose.position.x == 0.0:
tmp_box = BoundingBox()
tmp_box.header = box.header
tmp_box.pose = box.pose
tmp_box.dimensions = box.dimensions
# TODO fix index indent, jsk_pcl_ros_utils/label_to_cluster_point_indices_nodelet.cpp
tmp_box.label = index + 1
labeled_cluster_boxes.boxes.append(tmp_box)
for box, label in zip(instance_boxes_msg.boxes, instance_label_msg.labels):
tmp_box = BoundingBox()
tmp_box.header = box.header
tmp_box.pose = box.pose
tmp_box.dimensions = box.dimensions
tmp_box.label = label.id
labeled_instance_boxes.boxes.append(tmp_box)
self.labeled_cluster_boxes_pub.publish(labeled_cluster_boxes)
self.labeled_instance_boxes_pub.publish(labeled_instance_boxes)
def people_msg_callback(self, people, classes):
bboxes = BoundingBoxArray(header=people.header)
for p in people.poses:
b = BoundingBox()
for i, n in enumerate(p.limb_names):
if n in ["Neck", "Nose", "REye", "LEye", "REar", "LEar"]:
b.header = people.header
b.pose = p.poses[i]
break
if not b.header.frame_id:
b.header = people.header
b.pose = b.poses[0]
self.box_msg_callback(bboxes, classes)
def init_boundingboxarray(self, num_boxes=30):
self.boundingBoxArray_object = BoundingBoxArray()
h = std_msgs.msg.Header()
h.stamp = rospy.Time.now(
) # Note you need to call rospy.init_node() before this will work
h.frame_id = "world"
self.boundingBoxArray_object.header = h
self.minimum_dimension = 0.2
self.init_x_position = 1.0
for i in range(num_boxes):
new_box = BoundingBox()
new_box.header = h
new_box.pose = Pose()
new_box.pose.position.x = self.init_x_position + i * self.minimum_dimension
new_box.dimensions = Vector3()
new_box.dimensions.x = self.minimum_dimension
new_box.dimensions.y = self.minimum_dimension
new_box.dimensions.z = self.minimum_dimension
new_box.label = i
new_box.value = i * self.minimum_dimension
self.boundingBoxArray_object.boxes.append(new_box)
self.publish_once(self.boundingBoxArray_object)
def callback(self, box_msg):
labeled_boxes = {}
label_buf = []
orientation = Pose().orientation
for box in box_msg.boxes:
if box.label in label_buf:
labeled_boxes[box.label] += [self.get_points(box)]
orientation = box.pose.orientation
else:
labeled_boxes[box.label] = [self.get_points(box)]
label_buf.append(box.label)
bounding_box_msg = BoundingBoxArray()
for label, boxes in zip(labeled_boxes.keys(), labeled_boxes.values()):
thresh = self.thresh
if self.label_lst[label] == 'shelf_flont':
thresh = 2.0
clustering = Clustering()
boxes = np.array(boxes)
result = clustering.clustering_wrapper(boxes, thresh)
for cluster in result:
max_candidates = [
boxes[i][0] + (boxes[i][1] * 0.5) for i in cluster.indices
]
min_candidates = [
boxes[i][0] - (boxes[i][1] * 0.5) for i in cluster.indices
]
candidates = np.array(max_candidates + min_candidates)
dimension = candidates.max(axis=0) - candidates.min(axis=0)
center = candidates.min(axis=0) + (dimension * 0.5)
distances = self.get_distances(
np.array([boxes[i][0] for i in cluster.indices]), [
np.linalg.norm(boxes[i][1]) * 0.5
for i in cluster.indices
])
tmp_box = BoundingBox()
tmp_box.header = box_msg.header
tmp_box.dimensions.x = dimension[0]
tmp_box.dimensions.y = dimension[1]
tmp_box.dimensions.z = dimension[2]
tmp_box.pose.position.x = center[0]
tmp_box.pose.position.y = center[1]
tmp_box.pose.position.z = center[2]
tmp_box.pose.orientation = orientation
tmp_box.label = label
tmp_box.value = distances.mean()
bounding_box_msg.boxes.append(tmp_box)
bounding_box_msg.header = box_msg.header
self.box_pub.publish(bounding_box_msg)
def callback(msg):
box_array = BoundingBoxArray()
box_array.header = msg.header
for footstep in msg.footsteps:
box = BoundingBox()
box.header = msg.header
box.pose = footstep.pose
box.dimensions = footstep.dimensions
box_array.boxes.append(box)
pub.publish(box_array)
def callback(msg):
box_array = BoundingBoxArray()
box_array.header = msg.header
for footstep in msg.footsteps:
box = BoundingBox()
box.header = msg.header
box.pose = footstep.pose
box.dimensions = footstep.dimensions
box_array.boxes.append(box)
pub.publish(box_array)
def convert(header, xx_cube):
boundingboxarray_msg = BoundingBoxArray()
boundingboxarray_msg.header = header
num_xx = int(xx_cube.shape[0] / 8)
for i in range(num_xx):
cube_raw = xx_cube[range(8 * i, 8 * i + 8), :]
bb_in_camera = np.column_stack((cube_raw, np.ones((8, 1))))
bb_in_lidar = np.linalg.inv(calib.lidar_to_cam).dot(bb_in_camera.T).T
boundingbox_msg = BoundingBox()
boundingbox_msg.header = boundingboxarray_msg.header
# boundingbox中心点位置
boundingbox_msg.pose.position.x = bb_in_lidar[:, 0].mean()
boundingbox_msg.pose.position.y = bb_in_lidar[:, 1].mean()
boundingbox_msg.pose.position.z = bb_in_lidar[:, 2].mean()
# 寻找y坐标最小的顶点，计算相邻两个顶点的旋转角及边长
bb_bottom = bb_in_lidar[:4]
min_idx = np.where(bb_bottom[:, 1] == bb_bottom[:, 1].min())[0][0]
theta = math.atan2(
bb_bottom[(min_idx + 1) % 4, 1] - bb_bottom[min_idx, 1],
bb_bottom[(min_idx + 1) % 4, 0] - bb_bottom[min_idx, 0])
b_1 = (
(bb_bottom[(min_idx + 1) % 4, 1] - bb_bottom[min_idx, 1])**2 +
(bb_bottom[(min_idx + 1) % 4, 0] - bb_bottom[min_idx, 0])**2)**0.5
b_2 = (
(bb_bottom[(min_idx + 3) % 4, 1] - bb_bottom[min_idx, 1])**2 +
(bb_bottom[(min_idx + 3) % 4, 0] - bb_bottom[min_idx, 0])**2)**0.5
if theta < 90 * math.pi / 180:
rotation_angle = theta
dimension_x = b_1
dimension_y = b_2
else:
rotation_angle = theta - 90 * math.pi / 180
dimension_x = b_2
dimension_y = b_1
# boundingbox旋转角四元数
boundingbox_msg.pose.orientation.x = 0
boundingbox_msg.pose.orientation.y = 0
boundingbox_msg.pose.orientation.z = math.sin(0.5 * rotation_angle)
boundingbox_msg.pose.orientation.w = math.cos(0.5 * rotation_angle)
# boundingbox尺寸
boundingbox_msg.dimensions.x = dimension_x
boundingbox_msg.dimensions.y = dimension_y
boundingbox_msg.dimensions.z = bb_in_lidar[:, 2].max(
) - bb_in_lidar[:, 2].min()
boundingbox_msg.value = 0
boundingbox_msg.label = 0
boundingboxarray_msg.boxes.append(boundingbox_msg)
return boundingboxarray_msg
def get_box(msg):
box = BoundingBox()
get_box_pose_srv = rospy.ServiceProxy("/transformable_server_sample/get_pose", GetTransformableMarkerPose)
resp = get_box_pose_srv(target_name=msg.data)
box.pose = resp.pose_stamped.pose
box.header = resp.pose_stamped.header
get_box_dim_srv = rospy.ServiceProxy("/transformable_server_sample/get_dimensions", GetMarkerDimensions)
resp2 = get_box_dim_srv(target_name=msg.data)
box.dimensions.x = resp2.dimensions.x
box.dimensions.y = resp2.dimensions.y
box.dimensions.z = resp2.dimensions.z
return box
def callback(msg):
box_array = BoundingBoxArray()
box_array.header = msg.header
for footstep in msg.footsteps:
box = BoundingBox()
box.header = msg.header
box.pose = footstep.pose
box.dimensions = footstep.dimensions
box.pose.position.z += (z_max + z_min) / 2.0
box.dimensions.z = z_max - z_min
box_array.boxes.append(box)
pub.publish(box_array)
def callback(msg):
box_array = BoundingBoxArray()
box_array.header = msg.header
for footstep in msg.footsteps:
box = BoundingBox()
box.header = msg.header
box.pose = footstep.pose
box.dimensions = footstep.dimensions
box.pose.position.z += (z_max + z_min) / 2.0
box.dimensions.z = z_max - z_min
box_array.boxes.append(box)
pub.publish(box_array)
def labeled_pose_callback(self, pose_msg):
self.header = pose_msg.header
self.qatm_boxes = BoundingBoxArray()
self.qatm_boxes.header = pose_msg.header
for pose in pose_msg.poses:
tmp_box = BoundingBox()
tmp_box.header = pose_msg.header
tmp_box.pose = pose.pose
tmp_box.dimensions.x = 0.03
tmp_box.dimensions.y = 0.03
tmp_box.dimensions.z = 0.03
tmp_box.label = self.label_lst.index(pose.label)
self.qatm_boxes.boxes.append(tmp_box)
def callback(self, instance_boxes_msg, instance_label_msg):
labeled_instance_boxes = BoundingBoxArray()
labeled_instance_boxes.header = instance_boxes_msg.header
for box, label in zip(instance_boxes_msg.boxes, instance_label_msg.labels):
tmp_box = BoundingBox()
tmp_box.header = box.header
tmp_box.pose = box.pose
tmp_box.dimensions = box.dimensions
tmp_box.label = label.id
labeled_instance_boxes.boxes.append(tmp_box)
self.labeled_instance_boxes_pub.publish(labeled_instance_boxes)
def get_box(msg):
box = BoundingBox()
get_box_pose_srv = rospy.ServiceProxy(
"/transformable_server_sample/get_pose", GetTransformableMarkerPose)
resp = get_box_pose_srv(target_name=msg.data)
box.pose = resp.pose_stamped.pose
box.header = resp.pose_stamped.header
get_box_dim_srv = rospy.ServiceProxy(
"/transformable_server_sample/get_dimensions", GetMarkerDimensions)
resp2 = get_box_dim_srv(target_name=msg.data)
box.dimensions.x = resp2.dimensions.x
box.dimensions.y = resp2.dimensions.y
box.dimensions.z = resp2.dimensions.z
return box
def dummyBoundingBoxPublisher():
pub = rospy.Publisher('/dummy_bounding_box', BoundingBox, queue_size=1)
rospy.init_node('dummyBoundingBoxPublisher_node', anonymous=True)
rate = rospy.Rate(25)
boundingBox_object = BoundingBox()
i = 0
pose_object = Pose()
dimensions_object = Vector3()
minimum_dimension = 0.2
boundingBox_object.label = 1234
while not rospy.is_shutdown():
h = std_msgs.msg.Header()
h.stamp = rospy.Time.now(
) # Note you need to call rospy.init_node() before this will work
h.frame_id = "world"
boundingBox_object.header = h
sinus_value = math.sin(i / 10.0)
boundingBox_object.value = sinus_value
# Change Pose to see effects
pose_object.position.x = 1.0
pose_object.position.y = 0.0
pose_object.position.z = sinus_value
# ai, aj, ak == roll, pitch, yaw
quaternion = tf.transformations.quaternion_from_euler(ai=0,
aj=0,
ak=sinus_value)
pose_object.orientation.x = quaternion[0]
pose_object.orientation.y = quaternion[1]
pose_object.orientation.z = quaternion[2]
pose_object.orientation.w = quaternion[3]
dimensions_object.x = sinus_value / 10 + minimum_dimension
dimensions_object.y = minimum_dimension
dimensions_object.z = minimum_dimension
# Assign pose and dimension objects
boundingBox_object.pose = pose_object
boundingBox_object.dimensions = dimensions_object
pub.publish(boundingBox_object)
rate.sleep()
i += 1
def vis_callback(msg):
obj_bba = BoundingBoxArray()
obj_bba.header.frame_id = 'map'
for obj in msg.objects:
obj_bb = BoundingBox()
res = tosm.query_individual(obj.object_name + str(obj.ID))
obj_bb.header = obj.header
pose = res.pose[0].replace('[', '').replace(']', '').split(',')
size = res.size[0].replace('[', '').replace(']', '').split(',')
obj_bb.pose.position.x = float(pose[0])
obj_bb.pose.position.y = float(pose[1])
obj_bb.pose.position.z = float(pose[2]) + float(size[2]) / 2.0
obj_bb.pose.orientation.x = float(pose[3])
obj_bb.pose.orientation.y = float(pose[4])
obj_bb.pose.orientation.z = float(pose[5])
obj_bb.pose.orientation.w = float(pose[6])
# bounding box size
obj_bb.dimensions.x = float(size[0])
obj_bb.dimensions.y = float(size[1])
obj_bb.dimensions.z = float(size[2])
# likelihood
obj_bb.value = 1
# determine the color
if (obj.object_name == "hingeddoor"):
obj_bb.label = 1
elif (obj.object_name == "elevatordoor"):
obj_bb.label = 2
elif (obj.object_name == "A"):
obj_bb.label = 3
elif (obj.object_name == "B"):
obj_bb.label = 4
elif (obj.object_name == "C"):
obj_bb.label = 5
else:
obj_bb.label = 10
obj_bba.boxes.append(obj_bb)
ses_map_object_vis_pub.publish(obj_bba)
def inference_image(self, data):
try:
image = self.bridge.imgmsg_to_cv2(data, "bgr8")
except CvBridgeError as e:
print(e)
return
#if image captured
image_expanded = np.expand_dims(image, axis=0)
# Perform the actual detection by running the model with the image as input
(boxes, scores, classes,
num) = self.sess.run([
self.detection_boxes, self.detection_scores,
self.detection_classes, self.num_detections
],
feed_dict={self.image_tensor: image_expanded})
# Draw the results of the detection (aka 'visulaize the results')
vis_util.visualize_boxes_and_labels_on_image_array(
image,
np.squeeze(boxes),
np.squeeze(classes).astype(np.int32),
np.squeeze(scores),
self.category_index,
use_normalized_coordinates=True,
line_thickness=8,
min_score_thresh=0.9)
if len(boxes[0]) > 0 and scores[0][0] > 0.95:
[min_x, min_y, max_x, max_y] = boxes[0][0]
height, width = image.shape[:2]
# print(min_x * height, min_y * width, max_x * height, max_y * width)
# cv2.rectangle(image, (int(min_y * width), int(min_x * height)), (int(max_y * width), int(max_x * height)), (255, 0, 0), 2)
boundingbox = BoundingBox()
boundingbox.header = data.header
boundingbox.pose.position.x = int(min_y * width)
boundingbox.pose.position.y = int(min_x * height)
boundingbox.dimensions.x = int((max_y - min_y) * width)
boundingbox.dimensions.y = int((max_x - min_x) * height)
# print(boundingbox)
self.pub_boundingbox.publish(boundingbox)
# All the results have been drawn on image. Now display the image.
cv2.imshow('Object detector', image)
if cv2.waitKey(10) == ord('q'):
return
def candidateBoxes(header, model_index):
box_array = BoundingBoxArray()
box_array.header.stamp = header.stamp
box_array.header.frame_id = "odom"
dx = 0.1
for y in np.arange(0.0, 0.6, dx):
for z in np.arange(0.7, 1.0, dx):
for x in np.arange(0.0, -0.5, -dx):
box = BoundingBox()
box.header = box_array.header
box.pose.orientation.w = 1.0
box.pose.position.x = x
box.pose.position.y = y
box.pose.position.z = z
box.dimensions.x = 0.1
box.dimensions.y = 0.1
box.dimensions.z = 0.1
box_array.boxes.append(box)
return box_array
def LeadCarUpdate(msg):
global lidar_array, filter_boxes
leadc_boxes = BoundingBoxArray()
leadc_boxes.header = filter_boxes.header
compare_boxes = BoundingBoxArray()
compare_boxes.header = filter_boxes.header
compare_boxes.boxes = filter_boxes.boxes[:]
if len(msg.data)==0:
print('no lead car')
else:
rc_x = msg.data[0].pos_x
rc_y = msg.data[0].pos_y
rc_v = msg.data[0].speed
distances = []
for i in range(len(compare_boxes.boxes)):
lidar_x = compare_boxes.boxes[i].pose.position.x
lidar_y = compare_boxes.boxes[i].pose.position.y
distance = np.sqrt((rc_x-lidar_x)**2+(rc_y-lidar_y)**2)
distances.append(distance)
print(min(distances))
if (min(distances)<5):
obj_i = distances.index(min(distances))
leadc_boxes.boxes.append(compare_boxes.boxes[obj_i])
else:
leadc_box = BoundingBox()
leadc_box.header = filter_boxes.boxes[1].header
leadc_box.pose.position.x = rc_x
leadc_box.pose.position.y = rc_y
leadc_box.pose.position.z = .5
leadc_box.pose.orientation.x = 0
leadc_box.pose.orientation.y = 0
leadc_box.pose.orientation.z = 0
leadc_box.pose.orientation.w = 0
leadc_box.dimensions.x = 5
leadc_box.dimensions.y = 5
leadc_box.dimensions.z = 5
leadc_boxes.boxes.append(leadc_box)
pub_leadc.publish(leadc_boxes)
def candidateBoxes(header, model_index):
box_array = BoundingBoxArray()
box_array.header.stamp = header.stamp
box_array.header.frame_id = "odom"
dx = 0.1
for y in np.arange(0.0, 0.6, dx):
for z in np.arange(0.7, 1.0, dx):
for x in np.arange(0.0, -0.5, -dx):
box = BoundingBox()
box.header = box_array.header
box.pose.orientation.w = 1.0
box.pose.position.x = x
box.pose.position.y = y
box.pose.position.z = z
box.dimensions.x = 0.1
box.dimensions.y = 0.1
box.dimensions.z = 0.1
box_array.boxes.append(box)
return box_array
def prepareDetectionRegistration(self, centroid, now):
obj_det = BucketDetection()
obj_det.image = self.cv_bridge.cv2_to_imgmsg(self.stereo_left, "bgr8")
obj_det.tag = "object_tags/gate"
bbox_3d = BoundingBox()
bbox_3d.dimensions = Vector3(self.gate_dimensions[0], self.gate_dimensions[1], self.gate_dimensions[2])
bbox_pose = Pose()
x, y, z = list((np.squeeze(centroid)).T)
obj_det.position = Point(x, y, z)
bbox_pose.position = Point(x, y, z)
bbox_3d.pose = bbox_pose
bbox_header = Header()
bbox_header.frame_id = "duo3d_optical_link_front"
bbox_header.stamp = now
bbox_3d.header = bbox_header
obj_det.bbox_3d = bbox_3d
obj_det.header = Header()
obj_det.header.frame_id = bbox_header.frame_id
obj_det.header.stamp = now
return obj_det
def prepare_detection_registration(self, centroid, det, now):
obj_det = BucketDetection()
obj_det.image = self.cv_bridge.cv2_to_imgmsg(self.stereo_left, "bgr8")
obj_det.tag = str("object_tags/" + self.classes[det[0]])
bbox_dims = np.asarray([1.0, 1.0, 1.0])
if rospy.has_param("object_tags/" + self.classes[det[0]] + "/dimensions"):
bbox_dims = np.asarray(rospy.get_param("object_tags/" + self.classes[det[0]] + "/dimensions")).astype(float)
bbox_3d = BoundingBox()
bbox_3d.dimensions = Vector3(bbox_dims[0], bbox_dims[1], bbox_dims[2])
bbox_pose = Pose()
x, y, z = list((np.squeeze(centroid)).T)
obj_det.position = Point(x, y, z)
bbox_pose.position = Point(x, y, z)
bbox_3d.pose = bbox_pose
bbox_header = Header()
bbox_header.frame_id = "duo3d_optical_link_front"
bbox_header.stamp = now
bbox_3d.header = bbox_header
obj_det.bbox_3d = bbox_3d
obj_det.header = Header()
obj_det.header.frame_id = bbox_header.frame_id
obj_det.header.stamp = now
return obj_det
def talker():
pub = rospy.Publisher('/segmentation_decomposer/boxes',
BoundingBoxArray,
queue_size=10)
rospy.init_node('segmentation_talker', anonymous=True)
rate = rospy.Rate(10)
while not rospy.is_shutdown():
b = BoundingBoxArray()
b.header.frame_id = "/head_mount_kinect_rgb_optical_frame"
b.header.stamp = rospy.get_rostime()
b1 = BoundingBox()
# b1.header.frame_id = "/head_mount_kinect_rgb_optical_frame"
# b1.header.stamp = rospy.get_rostime()
b1.header = b.header
b1.pose.position.x = 0.1
b1.pose.position.y = 0.1
b1.pose.position.z = 0.1
b1.dimensions.x = 0.15
b1.dimensions.y = 0.06
b1.dimensions.z = 0.06
b.boxes = [b1]
pub.publish(b)
rate.sleep()
def callback(self, front_MarkerArray, back_MarkerArray, TwistStamped):
# print("front",len(front_MarkerArray.markers)/4)
# print("back",len(back_MarkerArray.markers)/4)
# # Concat front and back MarkerArray Messages
add_MarkerArray = copy.deepcopy(front_MarkerArray)
for i in range(len(back_MarkerArray.markers)):
add_MarkerArray.markers.append(back_MarkerArray.markers[i])
# print("add",len(add_MarkerArray.markers)/4)
# print("done")
if len(add_MarkerArray.markers) == 0:
return
header = add_MarkerArray.markers[0].header
frame = header.seq
boxes = BoundingBoxArray() #3D Boxes with JSK
boxes.header = header
texts = PictogramArray() #Labels with JSK
texts.header = header
obj_ori_arrows = MarkerArray() #arrow with visualization_msgs
velocity_markers = MarkerArray() #text with visualization_msgs
obj_path_markers = MarkerArray() # passed path
warning_line_markers = MarkerArray()
dets = np.zeros((0, 9)) # (None, 9) : 9는 사용할 3d bbox의 파라미터 개수
obj_box_info = np.empty((0, 7))
obj_label_info = np.empty((0, 2))
# frame을 rviz에 출력
overlayTxt = OverlayText()
overlayTxt.left = 10
overlayTxt.top = 10
overlayTxt.width = 1200
overlayTxt.height = 1200
overlayTxt.fg_color.a = 1.0
overlayTxt.fg_color.r = 1.0
overlayTxt.fg_color.g = 1.0
overlayTxt.fg_color.b = 1.0
overlayTxt.text_size = 12
overlayTxt.text = "Frame_seq : {}".format(frame)
det_boxes = BoundingBoxArray() #3D Boxes with JSK
det_boxes.header = header
# Receive each objects info in this frame
for object_info in add_MarkerArray.markers:
#extract info [ frame,type(label),tx,ty,tz,h,w,l,ry ]
if object_info.ns == "/detection/lidar_detector/box_markers":
tx = object_info.pose.position.x
ty = object_info.pose.position.y
tz = object_info.pose.position.z
l = object_info.scale.x
w = object_info.scale.y
h = object_info.scale.z
quaternion_xyzw = [object_info.pose.orientation.x, object_info.pose.orientation.y, \
object_info.pose.orientation.z, object_info.pose.orientation.w]
rz = tf.transformations.euler_from_quaternion(
quaternion_xyzw)[2]
obj_box_info = np.append(
obj_box_info,
[[-ty, -tz, tx - 0.27, h, w, l, -rz + np.pi / 2]],
axis=0)
size_det = Vector3(l, w, h)
det_box = BoundingBox()
det_box.header = header
det_box.pose.position = Point(tx, ty, tz)
q_det_box = tf.transformations.quaternion_from_euler(
0.0, 0.0, rz) # 어쩔 수 없이 끝단에서만 90도 돌림
det_box.pose.orientation = Quaternion(*q_det_box)
det_box.dimensions = size_det
det_boxes.boxes.append(det_box)
elif object_info.ns == "/detection/lidar_detector/label_markers":
label = object_info.text.strip()
if label == '':
label = 'None'
obj_label_info = np.append(obj_label_info, [[frame, label]],
axis=0)
dets = np.concatenate((obj_label_info, obj_box_info), axis=1)
self.pub_det_markerarray.publish(det_boxes)
del current_id_list[:]
# All Detection Info in one Frame
bboxinfo = dets[dets[:, 0] == str(frame),
2:9] # [ tx, ty, tz, h, w, l, rz ]
additional_info = dets[dets[:, 0] == str(frame), 0:2] # frame, labe
reorder = [3, 4, 5, 0, 1, 2,
6] # [tx,ty,tz,h,w,l,ry] -> [h,w,l,tx,ty,tz,theta]
reorder_back = [3, 4, 5, 0, 1, 2,
6] # [h,w,l,tx,ty,tz,theta] -> [tx,ty,tz,h,w,l,ry]
reorder2velo = [2, 0, 1, 3, 4, 5, 6]
bboxinfo = bboxinfo[:,
reorder] # reorder bboxinfo parameter [h,w,l,x,y,z,theta]
bboxinfo = bboxinfo.astype(np.float64)
dets_all = {'dets': bboxinfo, 'info': additional_info}
# ObjectTracking from Detection
trackers = self.mot_tracker.update(dets_all) # h,w,l,x,y,z,theta
trackers_bbox = trackers[:, 0:7]
trackers_info = trackers[:, 7:10] # id, frame, label
trackers_bbox = trackers_bbox[:,
reorder_back] # reorder_back bboxinfo parameter [tx,ty,tz,h,w,l,ry]
trackers_bbox = trackers_bbox[:,
reorder2velo] # reorder coordinate system cam to velo
trackers_bbox = trackers_bbox.astype(np.float64)
trackers_bbox[:, 0] = trackers_bbox[:, 0]
trackers_bbox[:, 1] = trackers_bbox[:, 1] * -1
trackers_bbox[:, 2] = trackers_bbox[:, 2] * -1
trackers_bbox[:, 6] = trackers_bbox[:, 6] * -1
# for문을 통해 각 objects들의 정보를 추출하여 사용
for b, info in zip(trackers_bbox, trackers_info):
bbox = BoundingBox()
bbox.header = header
# parameter 뽑기 [tx,ty,tz,h,w,l,rz]
tx_trk, ty_trk, tz_trk = float(b[0]), float(b[1]), float(b[2])
rz_trk = float(b[6])
size_trk = Vector3(float(b[5]), float(b[4]), float(b[3]))
obj_id = info[0]
label_trk = info[2]
bbox_color = colorCategory20(int(obj_id))
odom_mat = get_odom(self.tf2, "velo_link", "map")
xyz = np.array(b[:3]).reshape(1, -1)
points = np.array((0, 3), float)
if odom_mat is not None:
points = get_transformation(odom_mat, xyz)
# 이전 x frame 까지 지나온 points들을 저장하여 반환하는 함수
# obj_id와 bbox.label은 단지 type차이만 날뿐 같은 데이터
# path_points_list = points_path(tx_trk, ty_trk, tz_trk, obj_id)
path_points_list = points_path(points[0, 0], points[0, 1],
points[0, 2], obj_id)
map_header = copy.deepcopy(header)
map_header.frame_id = "/map"
bbox_color = colorCategory20(int(obj_id))
path_marker = Marker(
type=Marker.LINE_STRIP,
id=int(obj_id),
lifetime=rospy.Duration(0.5),
# pose=Pose(Point(0,0,0), Quaternion(0, 0, 0, 1)), # origin point position
scale=Vector3(0.1, 0.0, 0.0), # line width
header=map_header,
color=bbox_color)
path_marker.points = path_points_list
obj_path_markers.markers.append(path_marker)
# Tracker들의 BoundingBoxArray 설정
bbox.pose.position = Point(tx_trk, ty_trk, tz_trk / 2.0)
q_box = tf.transformations.quaternion_from_euler(
0.0, 0.0, rz_trk + np.pi / 2) # 어쩔 수 없이 끝단에서만 90도 돌림
bbox.pose.orientation = Quaternion(*q_box)
bbox.dimensions = size_trk
bbox.label = int(obj_id)
boxes.boxes.append(bbox)
picto_text = Pictogram()
picto_text.header = header
picto_text.mode = Pictogram.STRING_MODE
picto_text.pose.position = Point(tx_trk, ty_trk, -tz_trk)
# q = tf.transformations.quaternion_from_euler(0.7, 0.0, -0.7)
picto_text.pose.orientation = Quaternion(0.0, -0.5, 0.0, 0.5)
picto_text.size = 4
picto_text.color = std_msgs.msg.ColorRGBA(1, 1, 1, 1)
picto_text.character = label_trk + ' ' + str(bbox.label)
texts.pictograms.append(picto_text)
# GPS sensor values
oxtLinear = TwistStamped.twist.linear
# oxtLinear = TwistStamped.twist.linear
# Tracker들의 속도 추정
obj_velo, dx_t, dy_t, dz_t = obj_velocity([tx_trk, ty_trk, tz_trk],
bbox.label, oxtLinear)
if obj_velo != None:
obj_velo = np.round_(obj_velo, 1) # m/s
obj_velo = obj_velo * 3.6 # km/h
obj_velo_scale = convert_velo2scale(obj_velo)
# # Tracker들의 Orientation
q_ori = tf.transformations.quaternion_from_euler(
0.0, 0.0, rz_trk + np.pi / 2) # 어쩔 수 없이 끝단에서만 90도 돌림
obj_ori_arrow = Marker(
type=Marker.ARROW,
id=bbox.label,
lifetime=rospy.Duration(0.2),
pose=Pose(Point(tx_trk, ty_trk, tz_trk / 2.0),
Quaternion(*q_ori)),
scale=Vector3(obj_velo_scale, 0.5, 0.5),
header=header,
# color=ColorRGBA(0.0, 1.0, 0.0, 0.8))
color=bbox_color)
obj_ori_arrows.markers.append(obj_ori_arrow)
obj_velo_marker = Marker(type=Marker.TEXT_VIEW_FACING,
id=bbox.label,
lifetime=rospy.Duration(0.5),
pose=Pose(Point(tx_trk, ty_trk, tz_trk),
Quaternion(0.0, -0.5, 0.0,
0.5)),
scale=Vector3(1.5, 1.5, 1.5),
header=header,
color=ColorRGBA(1.0, 1.0, 1.0, 1.0),
text="{}km/h".format(obj_velo))
velocity_markers.markers.append(obj_velo_marker)
current_id_list.append(bbox.label)
# Warning object line
warning_line = Marker(
type=Marker.LINE_LIST,
id=int(obj_id),
lifetime=rospy.Duration(0.2),
pose=Pose(Point(0, 0, 0),
Quaternion(0, 0, 0, 1)), # origin point position
scale=Vector3(0.2, 0.0, 0.0), # line width
header=header,
color=ColorRGBA(1.0, 0.0, 0.0, 1.0))
d = dist_from_objBbox(tx_trk, ty_trk, tz_trk, size_trk.x,
size_trk.y, size_trk.z)
if d < MIN_WARNING_DIST:
warning_line.points = Point(tx_trk, ty_trk,
tz_trk), Point(0.0, 0.0, 0.0)
warning_line_markers.markers.append(warning_line)
# Change Outer Circle Color
outer_circle_color = ColorRGBA(1.0 * 25 / 255, 1.0, 0.0, 1.0)
if len(warning_line_markers.markers) > 0:
outer_circle_color = ColorRGBA(1.0 * 255 / 255, 1.0 * 0 / 255,
1.0 * 0 / 255, 1.0)
# ego_vehicle's warning boundary
outer_circle = Marker(
type=Marker.CYLINDER,
id=int(obj_id),
lifetime=rospy.Duration(0.5),
pose=Pose(Point(0.0, 0.0, -2.0), Quaternion(0, 0, 0, 1)),
scale=Vector3(8.0, 8.0, 0.1), # line width
header=header,
color=outer_circle_color)
inner_circle = Marker(
type=Marker.CYLINDER,
id=int(obj_id),
lifetime=rospy.Duration(0.5),
pose=Pose(Point(0.0, 0.0, -1.8), Quaternion(0, 0, 0, 1)),
scale=Vector3(7.0, 7.0, 0.2), # line width
header=header,
color=ColorRGBA(0.22, 0.22, 0.22, 1.0))
# ego-vehicle velocity
selfvelo = np.sqrt(oxtLinear.x**2 + oxtLinear.y**2 + oxtLinear.z**2)
selfvelo = np.round_(selfvelo, 1) # m/s
selfvelo = selfvelo * 3.6 # km/h
oxtAngular = TwistStamped.twist.angular
q_gps = tf.transformations.quaternion_from_euler(
oxtAngular.x, oxtAngular.y, oxtAngular.z)
# # ego-vehicle 사진 출력
ego_car = Marker(type=Marker.MESH_RESOURCE,
id=0,
lifetime=rospy.Duration(0.5),
pose=Pose(Point(0.0, 0.0, -1.8),
Quaternion(0, 0, 0, 1)),
scale=Vector3(1.5, 1.5, 1.5),
header=header,
action=Marker.ADD,
mesh_resource=CAR_DAE_PATH,
color=ColorRGBA(1.0, 1.0, 1.0, 1.0))
# Self ego Velocity
text_marker = Marker(type=Marker.TEXT_VIEW_FACING,
id=0,
lifetime=rospy.Duration(0.5),
pose=Pose(Point(-7.0, 0.0, 0.0),
Quaternion(0, 0, 0, 1)),
scale=Vector3(1.5, 1.5, 1.5),
header=header,
color=ColorRGBA(1.0, 1.0, 1.0, 1.0),
text="{}km/h".format(selfvelo))
for i in prior_trk_xyz.keys():
if i not in current_id_list:
prior_trk_xyz.pop(i)
self.pub_frame_seq.publish(overlayTxt)
self.pub_boxes.publish(boxes)
self.pub_pictograms.publish(texts)
self.pub_selfvelo_text.publish(text_marker)
# self.pub_selfveloDirection.publish(arrow_marker)
self.pub_objs_ori.publish(obj_ori_arrows)
self.pub_objs_velo.publish(velocity_markers)
self.pub_path.publish(obj_path_markers)
self.pub_warning_lines.publish(warning_line_markers)
self.pub_ego_outCircle.publish(outer_circle)
self.pub_ego_innerCircle.publish(inner_circle)
self.pub_ego_car.publish(ego_car)
