Exploiting synergies of mobile mapping sensors and deep learning for traffic sign recognition systems

This paper presents an efficient two-stage traffic sign recognition system. First, 3D point cloud data is acquired by a LINX Mobile Mapper system and processed to automatically detect traffic signs based on their retro-reflective material. Then, classification is carried out over the point cloud pro...

Descripción completa

Detalles Bibliográficos
Autores: Arcos García, Álvaro, Soilán, Mario, Álvarez García, Juan Antonio, Riveiro, Belén
Tipo de recurso: artículo
Estado:Versión publicada
Fecha de publicación:2017
País:España
Institución:Universidad de Sevilla (US)
Repositorio:idUS. Depósito de Investigación de la Universidad de Sevilla
OAI Identifier:oai:idus.us.es:11441/125642
Acceso en línea:https://hdl.handle.net/11441/125642
https://doi.org/10.1016/j.eswa.2017.07.042
Access Level:acceso abierto
Palabra clave:Mobile mapping sensors
Point cloud
Traffic sign
Deep learning
Convolutional neural network
Spatial transformer network
Descripción
Sumario:This paper presents an efficient two-stage traffic sign recognition system. First, 3D point cloud data is acquired by a LINX Mobile Mapper system and processed to automatically detect traffic signs based on their retro-reflective material. Then, classification is carried out over the point cloud projection on RGB images applying a Deep Neural Network which comprises convolutional and spatial transformer layers. This network is evaluated in three European traffic sign datasets. On the GTSRB, it outperforms previous state-of-the-art published works and achieves top-1 rank with an accuracy of 99.71%. Furthermore, a Spanish traffic sign recognition dataset is released.