Inspiration
I saw too many people dying or bed-ridden for months after a crash with their motorbikes or scooters. So i created a AI model so that it detects if rider is equipped with helmet.
What it does
It has a live camera feed, using python we put it into the model then it gives us the thresholds between two labels i.e.: helmet and nohelmet
How we built it
First i set up some images in a folder then i built a python code which trains the model using the set of images, later after the model started working i put a dataset from "MOBILENETV2BASE" (it is a trusted dataset page)with the image with and without helmets. It is a keras model.
Challenges we ran into
At first we thought of using in-built models from yolo, but we couldn't find the ones that suited our problem. So we came up with the idea of training our own modelWe could design the python code such that model's image processing array matched that of the webcams(rgb). This was my first project with AI models, so i didn't know which type of model i should use.
Accomplishments that we're proud of
The first time when i put a helmet and the model finally predicted the correct threshold.
What we learned
When we use CNN model we have always match the array set of the model to the camera feed.
What's next for ULTRASRMART-AI HELMET DETECTOR
To develop it in remote-small-computers so that we can monitor it real-time.
Where does it run
The model only runs where there is terminal available, because to run the code we need a few code lines to typed in the terminal.
To run the code: type in the terminal-
python webcam_classifier2.py webcam
To train the model:
python webcam_classifier.py train traindata/ valdata/
Log in or sign up for Devpost to join the conversation.