abhijitmahalle/drone_follow_me

Human Detection and Tracking using m500 quadrotor

Table of Contents

1. About The Project
   • Demo
2. Report
3. Getting Started
   • Prerequisites
   • Installation
4. Usage
5. Contributors
6. Contact
7. License

About The Project

This project deals with the implementation of human detection and tracking. The quadrotor used is ModalAI m500 model with VOXL Flight autonomy computer and PX4 flight controller.

Summary of tasks achieved:

• Implemented ROS node for accessing and viewing Yolo v5 output.
• Controlled yaw angle based on bounding box center.
• Setup 1D LiDAR for forward and backward motion.
• Programmed PD controllers for both motions.

(back to top)

Demo

Follow Me mode [Third Person View] (X2 Speed)

normal_third_person_view.online-video-cutter.com.1.mp4

Drone Camera Feed (X2 Speed)

normal_drone_camera_feed.online-video-cutter.com.mp4

• Edge Case - Multiple Person in frame.
• Edge Case - Person running.

(back to top)

Getting Started

These are the instructions to get started on the project. To get a local copy up and running follow these simple steps.

Prerequisites

• OS - Linux (tested)
• Object Detection model - Yolo v5
• Hardware - Hires Camera, 1D LiDAR
• Software - PX4, ROS 1, C++
• Package required - voxl_mpa_to_ros
• Docker - ROS melodic with opencv1.2 (mavros and mavros_extras required)
• Ground Station - Qgroundcontrol

Installation

1. Add the following inside bashrc of your local machine.

   export ROS_MASTER_URI=<Drone_IP>

2. Setup tflite service for yoloV5 and select hires camera.
3. Clone package human_tracking into src of workspace and catkin_make.

(back to top)

Usage

For starting offboard_mode. Open 4 terminals and do the following.

1. First terminal. ssh into drone and source ROS variables. Then enter following command to start voxl_mpa_to_ros

   roslaunch voxl_mpa_to_ros voxl_mpa_to_ros.launch

2. Second terminal. Get into docker image and source the workspace and run following to start controller.

   roslaunch human_tracking follower.launch

3. Third terminal. Get into docker image and source the workspace and run following to view camera output with bounding boxes.

   rosrun human_tracking view

4. Fourth terminal. Source .bashrc of local machine and run rqt and select following topic for image view.

   /human_bb_view

(back to top)

License

Distributed under the MIT License. See MIT for more information.

(back to top)