Skip to content

Latest commit

 

History

History
75 lines (56 loc) · 2.82 KB

README.md

File metadata and controls

75 lines (56 loc) · 2.82 KB
Raw

Drone-Eye

A drone project that performs object detection and make a search engine out of the drone feed. Project under Machine Learning and AI society of Developer Students Club - IIT Patna.

Motivation

Modern drones are be equipped with cameras and are very prospective for a variety of commercial uses such as aerial photography, surveillance, etc.n. However, there are more challenges with drones due to top-down view angles and real-time constraints. Additionally, a challenging problem is the strong weight and area constraint of embedded hardware that limits the drones to run computation intensive algorithms, such as deep learning, with limited hardware resource.

Aim

Drone-Eye is a framework that intends to tackle both problems while running on embedded systems that can be mounted onto drones.Deep neural networks, object detection and object searching are the three major components in our work.

Prerequisites

  1. Basic Deep Learning concepts.
  2. Basics of Convolutional Neural Networks.
  3. Object Detection using CNN
  4. Tools

Hardware Requirements

  1. Drone hexacopter or octacopter
  • PixHawk
  • PPM Encoder
  • Propellers
  • 6-8 Brushless Motors
  • 6-8 Electronic Speed Controllers
  • Landing Gear
  • Radio remote
  1. NVIDIA Jetson Nano
  2. Camera (USB Cam will do)
  3. Gimble

Pipeline

Stage 1

  1. Making/Getting a drone.
  2. Putting on camera.
  3. Computation board like Jetson.

Stage 2

  1. Apply YOLO and other models on the feed of the camera
  • Use object detection with web cam
  • Find how to do the same task on NVIDIA Jetson Nano
  • Config and Implement the processing on Jetson board with external camera
  • Deploy the setup on the drone
  • Get the relevant info from the drone

Stage 3

  1. Tag the detected objects with GPS location, Compass direction and Time stamp.
  2. Send these info on a server.
  3. Make a search engine on the detected objects.

Resources

http://www.deeplearningbook.org/

Further Works

Further we may deploy the model on a swarm of drones so that objects detected are not redundant.

Communication

Our chat channel is to be found on Discord here.

Contributors