Neural Network Traffic Signs Detection System Development

The convolutional neural networks as a modern approach represent great results in various tasks connected to image processing. Such field as computer vision is an appropriate place for neural networks utilization as an image information extractor. Simultaneously, autonomous cars require image proces...

Full description

Saved in:
Bibliographic Details
Published in2019 XXII International Conference on Soft Computing and Measurements (SCM) pp. 125 - 128
Main Authors Devyatkin, A. V., Filatov, D. M.
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.05.2019
Subjects
convolutional neural network
deep learning
traffic sign detection
Online AccessGet full text
DOI10.1109/SCM.2019.8903787

Cover

More Information
Summary:The convolutional neural networks as a modern approach represent great results in various tasks connected to image processing. Such field as computer vision is an appropriate place for neural networks utilization as an image information extractor. Simultaneously, autonomous cars require image processing system for traffic signs detection. The work represents convolutional neural network system for traffic signs detection. The "You Only Look Once V3" (YOLOv3) architecture is used with Keras software library. The system provides multiobject detection with confidence percentage and bounding box coverage. During training the system is accelerated on GPU with multithreaded data augmentation. The database used is German Traffic Signs Detection Benchmark that contains 900 images.
DOI:10.1109/SCM.2019.8903787