triton_object_recognition

Description

This package contains a ROS2 node that can use a YOLOv3 or YOLOv4 network to detect and classify objects within an image. Note that based on the config file passed to the object recognition node, it might take a while for the node to start for the first time because it will download the appropriate config and weight files to the path <PATH_TO_TRITON>/triton/install/triton_object_recognition/share/triton_object_recognition. These files tend to be quite large but they should only have to be downloaded once provided you do not delete the install folder.

Usage

You can use the following command to run the object_recognizer node using the yolov4-tiny model:

ros2 launch triton_object_recognition tiny_yolov4_launch.py

You should see Object Recognizer successfully started! if successful.

Nodes

• object_recognizer : A component node (triton_object_recognition::ObjectRecognizer) which recognizes objects in a received image.

  ### Subscribed Topics

  • object_recognizer/in (sensor_msgs/msg/Image.msg) : Input image.

  ### Published Topics

  • object_recognizer/out (triton_interfaces/msg/DetectionBoxArray.msg) : Output detection boxes.

  ### Services

  • object_recognizer/recognize (triton_interfaces/srv/ObjectDetection.srv): Takes in an image message and produces the detection boxes from that image.

Contributors

• Kevin Huang (kevinh42@student.ubc.ca)