TEAM DIJON

Project Helmet Detection is a project to detect riders on two wheelers without the helmet. This project is divided into different sections

OPTIONAL Sometimes the code may not run due to the difference in versions so kindly install the version we have given in the requirements.txt

First Section The first part is getting the dataset. We team dijon downloaded dataset from kaggle.The dataset has images and its annotations which is in xml file Our Train_XML file reads xml file and crops the images into two categories with helmet and non helmet Then it is given as input to mobileNet model and training is done

NOTE: TO RUN THE CODE. DOWNLOAD THE DATASET FROM HERE https://www.kaggle.com/andrewmvd/helmet-detection AND PUT 'images' folder and 'annotations' folder in same directory where Train_XMLfile.py is kept.

Second Section Now after the dataset is downloaded and trained helmet.h5 will be our trained model. A trained model is already provided in this repository.I have used 50 epochs with batchsize of 35.

If you want to see a demo run of helmet detection run the Project_Helmet_Dijon.py Make sure to keep the haarcascade_upperbody.xml and test.mp4 in the same directory

If you want to see demo from your webcam then run webcam_helmet_detect.py Make sure to keep the res10_300x300_ssd_iter_140000.caffemodel and deploy.prototxt.txt in same directory. These are used to detect face and feed the image to model that predicts the probability of wearing helmet.