High-Level Project Summary
We developed a tool consisting of 3 components : - Two DL models, one that is implemented in a drone in order to detect full trash cans with our scraped data , and the other one detects damaged roads.- An API that receives the predictions from the drones/satellites, processes that data and sends it in JSON format to a database.- A user friendly dashboard for both decision and management support, in which you can find the addresses sent by the satellites/drones , and their localisations in the map.Using this methodology, we solve the problem of municipalities' slowness to react to waste and infrastructure problems, while optimizing expenses. Thus, will have better living standards.
Link to Project "Demo"
Link to Final Project
Detailed Project Description
Our solutions help in collecting, analyzing and presenting the infrastructure and waste management indicators to serve for the SDG 11.
It consists of 3 main steps:
- Images collection by drones for waste management and by satellites for road maintenance (Data are collected using multiple open source resources such as AEB and ESA to train and test our ML models - Python, TensorFlow, openCV, vgg60)
- Using images from step1, and our image recognition deep learning model we predict the state of roads and garbage cans and send it to the UI using our own API (Django)
- Present a user-friendly dashboard for daily management and strategic decision support (Figma)
We want to achieve a safer roads and a cleaner cities, and in general the project is extensible for any other process (slum upgrading, public transport accessibility, green areas protection).
Space Agency Data
We used INPE's catalog to get satellites images data in order to work on our damaged roads detection. We manually labeled the data and used them for training. Unfortunately, we didn't have enough time to finish the training. We are willing to continue working on our algorithm for future competitions, until we get our final solution.
Hackathon Journey
The Space Apps experience was wonderful. My team and I got to manage our stress in order to optimize our time and workflow.
Together, we learned how to communicate and push each other. 24h of work isn't easy! We were obliged to alternate our work time and continuously communicate the progress to each other. The overall experience wasn't easy for us, but choosing the challenge that suites us the best wasn't complicated. We, in Tunisia, suffer from a bad quality of roads and a terrible garbage management system, that has a direct negative effect of Tunisians' daily lives. Consequently, we have fixed our goals early and started putting on our ideas on a white board, which helped us find our path quickly. After setting out our main objectives, we started searching the data needed, where we encountered some problems because of the different types of data we found. Thus, we decided to divide our team into two groups : a data modeling group and a design group. The data modeling group worked very hard to merge the data together, while the design group started working on the dashboard's and presentation's design.
My team and I wanted to thank GoMyCode startup for providing us with a free and comfortable co-working space where we've spent 24h working in peace.
References
Dataset collected by our team : - https://www.kaggle.com/mednoun/garbage-container-fullempty-classification
- https://biankatpas.github.io/Cracks-and-Potholes-in-Road-Images-Dataset/
Resources : -https://unstats.un.org/sdgs/metadata/?Text=&Goal=11&Target=
Tools : Open CV / Tensorflow / Figma / Google Colab
Tags
#drones #satellites #smart_city #MachineLearning#deeplearning#artificialIntelligence
Global Judging
This project has been submitted for consideration during the Judging process.