darknet-ROS integration for use in gazebo

Overview

This is a forked repository. This repo was used to perform object detection on the gazebo implementation of SAUVC'20 pool. A custom object detector was trained using YOLOv3 on darknet. Its corresponding weights and cfg file can be found in this repository.

Download this repository

git clone --recursive https://github.com/SubZer0811/darknet_ros
cd ../

Build

Using Catkin Command Line Tools:

catkin build darknet_ros -DCMAKE_BUILD_TYPE=Release

Setup

• camera_topic: change the topic at darknet_ros/config/ros.yaml (line 4)
• weight_path and cfg_path: darknet_ros/config/yolov3_auv_sim_gate_obstacle_v1.yaml is the file that contains the paths to cfg and weights file for darknet to use. weights and .cfg files need to placed at darknet_ros/darknet_ros/yolo_network_config/weights and darknet_ros/darknet_ros/yolo_network_config/cfg respectively.
• disabling frame output: toggle enable_opencv under imagee_view to enable and disable displaying the inferred frame in darknet_ros/config/ros.yaml (line 31)
• frame rate: Changing the rate (in Hz) value at ros.yaml will change the output frame rate of darknet_ros.

Run

roslaunch darknet_ros yolo_v3.launch

Actions

An image can be sent the action server /darknet_ros/check_for_objects/.

Documentation

Publishers

/darknet_ros/detection_image_WL: This topic publishes images that are inferred without labels.
/darknet_ros/detection_image: This topic publishes images that are inferred by the neural network. /darknet_ros/bounding_boxes: This topic publishes the detected objects with bounding boxes in yolo format. /darknet_ros/found_object: This topic publishes the sequence number, frame id and number of detected objects in the same frame.

Subscribers

/vec6/f_cam/image_raw: This is the topic that darknet_ros subscribes to recieve frames from the camera.

The queue size of every topic has been set to 1 to achieve a real time feedback.

config files

There are 2 configuration files (excluding the cfg file required by darknet) that are essential to run darknet_ros.

• darknet_ros/config/ros.yaml: This file contains the details of the various subscribers and publishers used. Setting the frame rate and frame output is done at this file.
• darknet_ros/config/<darknet_config_file>.yaml: This file holds the path to weights and cfg files required by darknet. The threshold of detection can be set here. This file also consists of all the classes used by the neural network.

Important points to keep in mind

The default method is to discard boundboxes with height and width less than 0.01. For now, this value has been set to 0.001.