object_detection_msgs

ROS message definitions for object detection in image

Messages

Objects

• stamp, names, probabilities, and contours of detected objects

Points

• contour points of a detected object

Point

• an image coordinate

Nodelet: SynchronizedObjectPublisher

• subscribe an image topic and publish user-specified objects synchronized to the image
• subscribe another object topic, merge it with user-specified objects and publish

Subscribed topics

image_raw (sensor_msgs/Image)

• an image topic to be synchronized
• set ~subscribe_image true to subscribe

objects_in (object_detection_msgs/Objects)

• an object topic to be merged and synchronized
• set ~subscribe_objects true to subscribe

Published topics

objects_out (object_detection_msgs/Objects)

• synchronized to an image or objects subscribed

Services

get_objects (object_detection_msgs/GetObjects)

• get user-specified objects

set_objects (object_detection_msgs/SetObjects)

• set user-specified objects

Parameters

~subscribe_image (bool, default: false)

• Either this or ~subscribe_objects must be true to let this nodelet work

~subscribe_objects (bool, default: false)

• Either this or ~subscribe_image must be true to let this nodelet work

~names (string[], default: <empty array>)

• names of published objects

~probabilities (double[], default: <empty array>)

• probabilities of published objects

~contours (int[][][2], default: <empty array>)

• contours of published objects

~image_transport (string, default: "raw")

• transport type for the subscribed image topic

Nodelet: NonMaximumSuppressor

• subscribe an object topic and suppress overlapped objects with lower probabilities

Subscribed topics

objects_in (object_detection_msgs/Objects)

Published topics

objects_out (object_detection_msgs/Objects)

Parameters

~score_threshold (double, default: 0.4)
~nms_threshold (double, default: 0.5)
~eta (double, default: 1.0)
~top_k (int, default: 0)

• parameters for non maximum suppression (NMS) algorithm
• see documentation of cv::dnn::NMSBoxes()

Examples

see launch/test_*.launch