Real-Time Number Plate and Helmet Detection of Motorcyclists using YOLOv5 and ResNet-50

Vehicles have played a major part in transportation in the modern world. Especially, in developing countries, Two-wheel vehicles such as Motorcycles have been more popular. Accidents involving motorcycles have become more frequent recently. One of the leading factors in fatal motorcycle accidents is the rider’s failure to wear a protective helmet. Traffic inspectors frequently observe motorcycle riders or review CCTV footage to make sure they are wearing helmets. Those who do not wear helmets are fined. However, it necessitates action and effort from people. This study suggests using CCTV data to automatically find motorcyclists who are not wearing helmets and get their license plates. Recently developed methods for detecting license plates have yielded encouraging results, some of them are: Cascade R-CNN, Mask R-CNN etc. Despite the fact that new methods for detecting license plates have produced encouraging results, there are still some drawbacks and restrictions attached to these methods. Here are a few of these drawbacks: Generalization, Computational Requirements, Dataset Bias, etc. The suggested method starts by removing the background from the video so that moving objects can be seen. Following that, moving items are categorized as motorcycle or non-motorcycle related. The rider’s head position is placed, further, it is categorized as a helmet or not. Lastly, the motorcycle’s number plate is found, and its characters are taken for any motorcyclist that fails to wear a helmet and has been identified by the system. The proposed system uses Resnet50, short for Residual Networks, a convolutional neural network with 50 layers that uses transfer learning that is put forth on pre-trained models for classification that paved the way to the final detection of motorcycle helmets has a high accuracy (mAP) of 98.89%, an F1-score of 94.6, a detection speed of 130 frames per second, and making it more reliable.