ROS Node

DetectionSuite Deployer functionality is provided as a ROS Node. This node uses some input images stream (it could be a video) and makes inferences over the images passed, detecting the objects in realtime. Currently it only supports tensorflow as inferencer, but a broader support is expected in the future.

To use DetectionSuite ROS Node, it needs to subscribe to a rostopic. It would publish images that the node would infer.

In the build/devel directory inside DetectionSuite folder, execute:

source setup.bash

While the video is playing, start the ros node with the following command:

rosrun detection_suite test _topic:=<topic name> _configfile:=<path/to/configfile>

_topic is the topic detection suite is suscribed to and _confilefile the configuration file with the inferencer, class names… The detections are publishes at my_topic. Run the following command to see available topics:

rostopic list

To check the detections finally run the following command that should list the objects detected with their position in the image and the confidence for the detection.

rostopic echo /my_topic

Use example

In the example, the node is subscribed to a stream using video_stream_opencv ROS driver. In this case, the only needed file is video.launch, where the local video file should be set, so it detects it. After this stream is launched using

roslaunch video_stream_opencv video_file.launch

a window should pop up with the video.