Scribd
Upload
9 views5 pages

Aerothon ROS2 Simulation Report

The Aerothon ROS Report details a simulation setup using Ubuntu 22.04 and ROS2 Humble with Gazebo-Classic for drone operations. It outlines the functionality of various ROS nodes, including object detection via YOLO and the transformation of 2D bounding box data into 3D world coordinates for path planning. The report also describes the process of averaging nearby detected points to enhance target tracking accuracy.

Uploaded by

luvkushaa
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
9 views5 pages

Aerothon ROS2 Simulation Report

The Aerothon ROS Report details a simulation setup using Ubuntu 22.04 and ROS2 Humble with Gazebo-Classic for drone operations. It outlines the functionality of various ROS nodes, including object detection via YOLO and the transformation of 2D bounding box data into 3D world coordinates for path planning. The report also describes the process of averaging nearby detected points to enhance target tracking accuracy.

Uploaded by

luvkushaa
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Aerothon ROS Report

Simulation Specifications:
OS: Ubuntu Jammy Jellyfish (22.04)
ROS2 distro: Humble
Simulation software: Gazebo-Classic
Model: Iris_vision.sdf
World: custom_world.world
Control Mode: OFFBOARD
Bridge: MAVROS (MAVLINK Integration)

ROS Nodes:
1. PX4_sitl node: Launches the PX4 simulation with Gazebo.
2. YOLO Node: Applies YOLOv10n on drone’s camera feed to detect objects and publish their
centers.
3. MAVLINK controller: Interfaces MAVROS with PX4, arms drone, and sends target commands
via /target_point.
4. Path Planner node: Converts 2D bounding box coordinates into 3D world coordinates and sends
movement commands.
Mapping YOLO coordinates to Real-World Coordinates
Mapping YOLO coordinates to Real-World Coordinates
The path_planner.py node handles the transformation of 2D YOLO bounding box data into
actionable 3D world coordinates. Here’s the process in detail:

1. Input
Bounding box center coordinates (x_center, y_center) received on /bbox_center.
Camera intrinsics:
Focal length along X and Y axes: fx and fy
Coordinates of center of video frame: cx and cy

2. Camera Intrinsics Normalization


The pinhole camera model is used to normalize 2D points:

This produces a 3D ray direction in the camera frame (ray_cam = [x_norm, y_norm, 1.0]).

3. Obtaining “camera_link” transform frame


The camera_link transform relative to base_link is obtained via a static transform published by the
static_transform_publisher node in my launch file. This defines the camera's fixed
position/orientation on the drone:

The “camera_link” is defined 0.1m below the


base_link of the drone.
4. TF2 Transform lookup
The latest transform from “camera_link” to “map” is retrieved using this:

5. Rotation Matrix from Quaternion (Math Foundation)


The rotation matrix R is calculated from a quaternion q = (x, y, z, w) using the following formula:

Quaternion Normalization Ensures proper rotation representation.

6. Ray Transformation to World frame


The camera ray direction is rotated into the world frame:

7. Averaging Nearby points


To handle repeated detections in the same region: As the drone moves, the YOLO model may detect
the same object multiple times from slightly different angles or positions.
• Each new detection is compared with previously recorded points using Euclidean distance.
• If the new point is within a minimum distance threshold (min_dist_threshold), it is
considered to be part of the same target.
• Instead of storing each close detection separately, the system treats it as a duplicate of an
already known point.
• The coordinates of nearby points are averaged using a moving average.
• Maintains historical points for path planning
Final Simulation

You might also like