Crowd Informer

Introduction

Life was normal until the COVID-19 pandemic. In addition to posing a high risk on physical health, the pandemic also affected global mental health, and has affected everyone's day to day lives. Global economy has slowed down due to this pandemic and almost everything has become difficult for everyone, especially those in high risk areas.

Containing the spread of virus has been an international priority. Many countries have suffered from the first wave of this pandemic and lockdowns were one way of containing the spread in high risk areas. People are advised to wear masks, practice hand hygiene and social distancing rules. Another way to protect yourself from coronavirus is to avoid crowds.

In my country, Sri Lanka, the local police is strictly enforcing masks and social distancing violators and punishes violators while other anti-virus measures, such as lockdowns, are eased to reduce the economic pain.

Crowding should be avoided during this pandemic. Even though people are aware of this, they may forget and crowd in certain places, increasing the risk of infection. To prevent this, I came up with the idea of building an autonomous drone system to identify crowded places and share the location with the local police.

Hardware Components

• HoverGames Drone Developer Kit (with Remote Controller)
• Telemetry radio
• Battery and charger : 3S 3500-5000mAh XT60 LiPo
• Raspberry Pi 3 Model B
• Raspberry Pi Camera module
• Intel Neural Compute Stick 2

Software and OS

• PX4 Autopilot
• Raspberry Pi Raspbian
• Intel OpenVINO toolkit

Intel Neural Compute Stick 2 and OpenVINO Toolkit

For this project, I will be using the Intel Neural Compute Stick 2 and OpenVINO toolkit to help my autonomous drone system detect crowding. The Intel Neural Compute Stick 2 features the Intel Movidius Myriad X VPU which is a self-sufficient, all-in-one processor that features the powerful Neural Program Engine and 16 SHAVE cores that deliver class-leading performance for deep neural network interference applications.

The Intel Neural Compute Stick 2 running OpenVINO will be connected to the Raspberry Pi 3.

You can follow this guide to install OpenVINO on Raspberry platform to use with the Intel Neural Compute Stick 2.

Crowd detection with OpenVINO

I will be using OpenVINO's pre-trained object detection model to develop a deep learning software to detect people. We can make a slight change in the codes to detect crowds. If the number of people detected is high, then our drone will identify the group as a crowd.

The Pre-trained Machine learning model can be trained to learn the human body variabilities in order to detect humans in video streams.

Communication between FMU and Raspberry Pi

Please follow this helpful guide to learn how to make your FMU communicate with your Raspberry Pi.

GPS

The RDDRONE-FMUK66 flight management unit is the foundation for building industrial robotic drones, rovers, and other small autonomous vehicles. This reference design runs PX4, the standard for industrial-grade drones, and gives you freedom to develop your own robotic vehicle.

It controls and directs the vehicle's navigation and real-time response to its environment. It is adaptable to many airframes and vehicle types, including ground and water-based robots. It performs sensor fusion, including GPS and other positioning inputs for autonomous navigation to mission way points. The open, extensible platform supports many additional sensors.

A GSM module will be connected to the Raspberry Pi so that as soon as the crowd is detected, the GPS location will be recorded and shared with the local police.

Future work

This autonomous drone system can be used to monitor crowds and alert the police. This system will be useful even after the pandemic and could be used in many ways, such as identifying overcrowding and alerting people about the crowds.