Automatic Detection of Helmet and License Plate Recognition using CNN & GAN

Abstract

Enforcing use of helmet on every bike-rider is mandatory nowadays due to high accident rate and
poor conditioned roads. There are laws regarding safety measures which ensure use of helmet. But
currently, they involve manual intervention which is not proven to be so effective because sometimes,
bike-riders tend to escape without any penalty after breaking the safety rules like wearing a helmet
while riding. Automation is efficient & also a better way to deal with this problem but it comes with
its own challenges. To name a few, Low quality image frames (low image resolution, pixel density
etc.), rain, dew & fog and partly hidden faces. Hence, the robustness of detection methodology
strongly depends on the strength of extracted features and also the ability to deal with the lower
quality of extracted data. The first goal of this project is to boost the potency of helmet detection and
then recognizing the license number plate recognition. This model consists of many essential steps
developed using today’s most advanced & optimized Convolutional Neural Network (CNN),
Generative Adverbial Network (GAN) models & libraries. This model is a classification based model
that uses supervised learning approach to train CNN and GAN. The proposed helmet detection model
can be used to detect helmet and recognizes license plate even in adverse conditions.