As more vehicles ply on roads around the world, there is a rapid increase in the number of road incidents. Currently traffic police and cameras are deployed at traffic signals and other select spots, leaving room for large stretches of blind spots. Moreover, it is difficult to record the evidence of violations before the vehicle speeds away. Hence, this paper proposes a solution to leverage the front mounted cameras on the car's windshield connected to an onboard Raspberry Pi to photograph, record, and autonomously report traffic violations and upload them to a centralized database from anywhere on the road. Vehicles can automatically identify and report traffic violations around them. A modified You Look Only Once v3 model is proposed to detect crashes and helmet violations. The model achieves an accuracy of 88% and a lesser forward propagation time (500 ms). The model is implemented on Raspberry Pi using TensorFlow Lite to enable real-time implementation.
Crash and Helmet Violation Detection System Using On-Device YOLOv3 Model
2024-05-16
1129557 byte
Conference paper
Electronic Resource
English
Smart Helmet with Crash Detection
Springer Verlag | 2024
|Non-motor vehicle helmet wearing detection method based on improved YOLOv3 algorithm
European Patent Office | 2021
|Traffic Monitoring System using YOLOv3 Model
IEEE | 2023
|Helmet violation processing using deep learning
IEEE | 2018
|Smart Parking System Using YOLOv3 Deep Learning Model
Springer Verlag | 2023
|