It's a holiday and you decided to go on long drive but you wonder what will you do if some animal smashes your car? You definitely don't want to start your day by injuring a life. Quoting Wikipedia

Very large numbers of mammals, birds, reptiles, amphibians and invertebrates are killed on the world's roads every day. The number of animals killed in the United States has been estimated at a million per day. Road kill is major issue. It is driving some species towards extinction. So along with other issues like climate change or global warming this problem also needs to be addressed. I wondered what we can do to tackle this? So here is solution to the problem

What it does

My project comprises of two things.

  1. First is Mobile app
  2. And second is a python script which uses custom trained yolov6 model
  • The script can be used with car's dash camera which would perform detection in real time.
  • As soon as a animal is detected a message will be send on user's phone number using Twilio.
  • So user can check in real time and control his/her vehicle to stop accidents.
  • Using mobile app the user can check if they are in accident prone zone or not.
  • App detects the location of user and shows risk and also recommends them speed limit.
  • Thus user can drive his/her car peacefully and can effectively solve roadkill problem

How we built it

  • I have trained about 200 images (5 types of animals namely Florida panther, Badger, Koala, Wolf and squirrel) which I collected through google and trained for 10 epochs.
  • This is used in my Open CV script to check if any animal is present in image.
  • This script also sends image and text message on whatsapp using twilio.
  • The mobile app is created using Flutter.
  • I am using various widgets like location detection to detect live widget and maps to plot a map.

Challenges we ran into

  • This was my first time training a custom model using yolo so I had hard time figuring out how things work. I should have trained my model for some more time to get accurate results.
  • Also I wished to apply multithreading to make detection process much faster as opencv detects image frame by frame.

What's next for Shoot to thrill pray not to kill

A more faster and accurate version

Built With

Share this project: