Object Detection

Here are shown ros_openvino subscribed topics, published topics and parameters for object_detection node.

Subscribed topics

• /object_detection/input_image (sensor_msgs/Image)

  input image for RGB analysis

• /object_detection/input_depth (sensor_msgs/Image)

  input depth for RGBD analysis, depth_analysis parameter can disable this subscriber

• /object_detection/camera_info (sensor_msgs/CameraInfo)

  input camera correction for RGBD analysis, depth_analysis parameter can disable this subscriber

Published topics

• /object_detection/output_image (sensor_msgs/Image)

  output image, output_as_image can disable this publisher

• /object_detection/results (ros_openvino/Objects)

  object detection results as a list, output_as_list can disable this publisher

• /object_detection/markers (visualization_msgs/MarkerArray)

  object detection results as rviz markers, output_as_markers and output_as_markerslabel can disable this publisher

• /object_detection/box_list (ros_openvino/ObjectBoxList)

  object detection results as list of 3D boxes, output_as_box:list can disable this publisher

Parameters

• ~target (string, default: "MYRIAD")

  target device MYRIAD, GPU, CPU

• ~threshold (float, default: "0.5")

  confidence threshold to accept result

• ~model_network (string, default: "/opt/intel/computer_vision_sdk/deployment_tools/model_downloader/object_detection/common/mobilenet-ssd/caffe/mobilenet-ssd.xml")

  it is path to network model (a .xml file), you can use file in models folder

• ~model_weights (string, default: "/opt/intel/computer_vision_sdk/deployment_tools/model_downloader/object_detection/common/mobilenet-ssd/caffe/mobilenet-ssd.bin")

  it is path to weights for used model (a .bin file), you can use file in models folder

• ~model_labels (string, default: "/opt/intel/computer_vision_sdk/deployment_tools/model_downloader/object_detection/common/mobilenet-ssd/caffe/mobilenet-ssd.labels")

  it is path to a text file containing results labels (a .labels file), you can use file in models folder.

• ~model_colors (string, default: "")

  it is path to a text file containing results colors (a .colors file), you can use file in models folder.

• ~output_as_image (bool, default: "true")

  output as an sensor_msgs/Image message.

• ~output_as_list (bool, default: "true")

  output as a ros_openvino/Objects message

• ~frame_id (string, default: "/camera_link")

  frame_id used for all measures

• ~depth_analysis (bool, default: "true")

  enable depth analysis, some subscribers, publisher and parameteres won't be available if this mode is disabled

• ~output_as_markers (bool, default: "true")

  if depth_analysis is true, show cubes in detected area in rviz as visualization_msgs/MarkerArray message

• ~output_as_markers (bool, default: "true")

  if depth_analysis is true, show cubes in rviz as visualization_msgs/MarkerArray

• ~output_markers_lifetime (float, default: "0.1")

  if depth_analysis and output_as_markers or output_as_markerslabel are true, set markers lifetime in rviz in seconds

• ~output_as_box_list (bool, default: "true")

  if depth_analysis is true, show results as ros_openvino/ObjectBoxList message

Models

Models are not mine but crafted by me using OpenVINO model converter. Models are downloaded using Intel OpenVINO and so are under their licenses.
Parameters used for MYRIAD are:

• --data_type FP16
• --input_model mobilenet-ssd.caffemodel
• --input proto mobilenet-ssd.prototxt
• --mean_values [127.5,127.5,127.5]
• --scale 255
  and for GPU
  
• --data_type FP32
• --input_model mobilenet-ssd.caffemodel
• --input_proto mobilenet-ssd.prototxt