All over the world, the number of vehicles stolen is enormous. In Mexico, the number of cars stolen each year has increased every year for the past 5 years. It is becoming quite dangerous to own a car and getting it back is becoming harder. The insurance companies are increasing their prices and is becoming very expensive to own certain types of cars. Even freight trucks are being robbed and it is said that in certain localities even the police is involved in the crime.
The amount of cars in Mexico is close to 30Million. It is becoming quite common in some cities to install speed radar cameras in mayor streets but the government has not realized that with this infrastructure already in place and working they could use it to also monitor and spot recently stolen vehicles.
I love what the cripto-economy has started to do to our world by applying some basic principles of game theory. I realized that with the same principles of game theory and thinking in ways to not fall into "the tragedy of the commons" with this particular problem, we could use technology to start better watching for ourselves.
I have counted that for every 1minute driving, I pass over at least +25 different cars (parked, same lane, opposite lane, cross lane etc...). Imagine that 10% or maybe a lot less of the total vehicles in Mexico had eyes and could read every license plate it could non-stop (even when the car is parked) and as soon as someone reported a stolen car, the system could anonymously send a location with a timestamp cc to the state police, news media, all insurance companies and all members of the social platform.
What it does
The system is just a camera connected to a board (tessel2, arduino, raspeberry Pi.. any board) with a gps sensor and a Gsm chip to constantly send license plates numbers to the main system looking for reported stolen registries.
Once the system detects the stolen car, the user that had the camera installed WONT know it has found a stolen car.
How I built it
The main challenge was to be able to detect license plates and the license plate numbers. I use a CNN (Convolutional neural network called Yolo3) to do the image processing and it is mounted on a google cloud server.
I used the tessel2 board since I found that it is really easy and fast to program it.
The main server is hosted in heroku and I use the gcloud server only as a processing service.
Facebook messenger is used to report the stolen car
Challenges I ran into
Train the neural network Structuring the idea to make it as transparent as possible Privacy is the main concern which needs still much thinking to preserve it Not having enough time to develop the idea
Accomplishments that I'm proud of
Trainning the neural network with very few real pictures Being able to use low budget cameras and achieve good confidence on the detection
What I learned
It is hard to design a fully transparent but highly private systems to a social purpose.
What's next for Friends Watching
Wait for the contest to finish and receive feedback.