This repo contains the project submitted for the Digital Image Processing course. The ultimate goal of this project is detecting whether people in video do not violate the 2m distancing rule by using computer vision algorithms. One of the requirements of this project was to use deep learning or machine learning as well, thus we used YOLO. We forked the repository of TrainYourOwnYOLO and built our project on top of it. Below, you can view all the information as provided in the README file of TrainYourOwnYOLO
Trivial
- Highest graded project 🏆
Prerequistics
In order for us to train our network we used the following databases of images:
- Penn-Fudan Database (~200 images)
- INRIAPerson (~100 images)
- Images we found on the internet (~50 images) of pedestrians, people on bikes etc.
- Our images that we took with a drone that involves people in the campus
Images in 3rd and 4th bullet points had to be manually annotated where we used the open-source tool labelImg.
We used around a random 20% as Test-data, and from the rest 80% we used 90% as Train-data and 10% as Validatiion-data