Drones and Artificial Intelligence to enforce social isolation during COVID-19 outbreak

Original article can be found here (source): Artificial Intelligence on Medium

Drones and Artificial Intelligence to enforce social isolation during COVID-19 outbreak

Due to the lack of a vaccine so far one of the most effective recommendations adopted by multiple governments in the countries affected by COVID-19 is social distancing or isolation in order to “flatten the curve”, this means the spread can be slowed by avoiding public spaces and crowded places, reaching a lower amount of new cases in a specific time frame, a slower infection rate means a less stressed health care system, fewer hospital visits on any given day and fewer sick people being turned away.

A sample epidemic curve, with and without social distancing. (Image credit: Johannes Kalliauer/ CC BY-SA 4.0)

The terms for mandatory social isolation in some countries (I live in Lima, Perú for example) are staying at home unless you must go out under special conditions such as restocking food, family or medical emergencies. However, enforcing this measure involves a large number of police and military officers on the street patrolling 24/7 for the duration of the quarantine and considering that it could be extended according to the evolution in the contagion rate this could be a quite high effort on time and cost, in addition to the risk of contagion from the personnel patrolling the streets.

An idea so we could efficiently reinforce these measures is by combining different artificial intelligence techniques such as image processing and classification (computer vision) captured by drones. A drone is an unmanned vehicle that can be remotely operated or configured to follow a predefined route, in this way we could capture information on the streets safely, reducing the probability of contagion by official personnel, as an example:

After applying image recognition algorithms to video and getting present objects we can structure this information and use it to analyze the volume of people, cars, bikes or others circulating on streets and avenues. Having this information now structured would allow us to define strategies and plan.

Data Visualization of detected People, Cars and Bikes by District. *Important: the figures shown above are for demonstration purposes only, are not real and are not intended to represent reality.

Using this information we can achieve some of the following:
1) Optimize the deployment of patrols in charge of the streets and avenues
2) Minimizes 24/7 surveillance operating costs
3) Reduce the risk of contagion for walking patrol
4) Cover the less accessible areas due to the presence of contagion or high level of crime
5) Join data between non-compliance zones with social isolation measures the number of new cases

We must confront this pandemic with the new weapons that this generation has, one of them perhaps the most important today is the vast amount of data we can generate with technology and the available algorithms that gives life to artificial intelligence systems.

Streets of Lima Center with OpenCV (Original Video: https://www.youtube.com/watch?v=0myE-TDPqlI)

Technical details:

  1. The 1st video was recorded with a Tello Drone
  2. I used a YoloV2 architecture for the image recognition and pretrained weights with COCO dataset
  3. Folium for the map

Also… while learning don’t forget to have fun…

My own trained model to find Batman!

Related links: