Traffic signal detection and classification in street views using an attention model

Peter Hall, Yifan Lu, Jiaming Lu, Songhai Zhang

Research output: Contribution to journalArticle

Abstract

Detecting small objects is a challenging task. We focus on a special case: the detection and classification of traffic signals in street views. We present a novel framework that utilizes a visual attention model to make detection more efficient, without loss of accuracy, and which generalizes. The attention model is designed to generate a small set of candidate regions at a suitable scale so that small targets can be better located and classified. In order to evaluate our method in the context of traffic signal detection, we have built a traffic light benchmark with over 15,000 traffic light instances, based on Tencent street view panoramas. We have tested our method both on the dataset we have built and the Tsinghua–Tencent 100K (TT100K) traffic sign benchmark. Experiments show that our method has superior detection performance and is quicker than the general faster RCNN object detection framework on both datasets. It is competitive with state-of-the-art specialist traffic sign detectors on TT100K, but is an order of magnitude faster. To show generality, we tested it on the LISA dataset without tuning, and obtained an average precision in excess of 90%.

LanguageEnglish
Pages253-266
Number of pages14
JournalComputational Visual Media
Volume4
Issue number3
Early online date4 Aug 2018
DOIs
StatusPublished - 1 Sep 2018

Keywords

  • CNN
  • small object detection
  • traffic light benchmark
  • traffic light detection

ASJC Scopus subject areas

  • Computer Graphics and Computer-Aided Design
  • Computer Vision and Pattern Recognition
  • Artificial Intelligence

Cite this

Traffic signal detection and classification in street views using an attention model. / Hall, Peter; Lu, Yifan; Lu, Jiaming; Zhang, Songhai.

In: Computational Visual Media, Vol. 4, No. 3, 01.09.2018, p. 253-266.

Research output: Contribution to journalArticle

Hall, Peter ; Lu, Yifan ; Lu, Jiaming ; Zhang, Songhai. / Traffic signal detection and classification in street views using an attention model. In: Computational Visual Media. 2018 ; Vol. 4, No. 3. pp. 253-266.
@article{82a0457785f84461b64245ed2b9ddc7e,
title = "Traffic signal detection and classification in street views using an attention model",
abstract = "Detecting small objects is a challenging task. We focus on a special case: the detection and classification of traffic signals in street views. We present a novel framework that utilizes a visual attention model to make detection more efficient, without loss of accuracy, and which generalizes. The attention model is designed to generate a small set of candidate regions at a suitable scale so that small targets can be better located and classified. In order to evaluate our method in the context of traffic signal detection, we have built a traffic light benchmark with over 15,000 traffic light instances, based on Tencent street view panoramas. We have tested our method both on the dataset we have built and the Tsinghua–Tencent 100K (TT100K) traffic sign benchmark. Experiments show that our method has superior detection performance and is quicker than the general faster RCNN object detection framework on both datasets. It is competitive with state-of-the-art specialist traffic sign detectors on TT100K, but is an order of magnitude faster. To show generality, we tested it on the LISA dataset without tuning, and obtained an average precision in excess of 90{\%}.",
keywords = "CNN, small object detection, traffic light benchmark, traffic light detection",
author = "Peter Hall and Yifan Lu and Jiaming Lu and Songhai Zhang",
year = "2018",
month = "9",
day = "1",
doi = "10.1007/s41095-018-0116-x",
language = "English",
volume = "4",
pages = "253--266",
journal = "Computational Visual Media",
issn = "2096-0433",
publisher = "Tsinghua University Press",
number = "3",

}

TY - JOUR

T1 - Traffic signal detection and classification in street views using an attention model

AU - Hall, Peter

AU - Lu, Yifan

AU - Lu, Jiaming

AU - Zhang, Songhai

PY - 2018/9/1

Y1 - 2018/9/1

N2 - Detecting small objects is a challenging task. We focus on a special case: the detection and classification of traffic signals in street views. We present a novel framework that utilizes a visual attention model to make detection more efficient, without loss of accuracy, and which generalizes. The attention model is designed to generate a small set of candidate regions at a suitable scale so that small targets can be better located and classified. In order to evaluate our method in the context of traffic signal detection, we have built a traffic light benchmark with over 15,000 traffic light instances, based on Tencent street view panoramas. We have tested our method both on the dataset we have built and the Tsinghua–Tencent 100K (TT100K) traffic sign benchmark. Experiments show that our method has superior detection performance and is quicker than the general faster RCNN object detection framework on both datasets. It is competitive with state-of-the-art specialist traffic sign detectors on TT100K, but is an order of magnitude faster. To show generality, we tested it on the LISA dataset without tuning, and obtained an average precision in excess of 90%.

AB - Detecting small objects is a challenging task. We focus on a special case: the detection and classification of traffic signals in street views. We present a novel framework that utilizes a visual attention model to make detection more efficient, without loss of accuracy, and which generalizes. The attention model is designed to generate a small set of candidate regions at a suitable scale so that small targets can be better located and classified. In order to evaluate our method in the context of traffic signal detection, we have built a traffic light benchmark with over 15,000 traffic light instances, based on Tencent street view panoramas. We have tested our method both on the dataset we have built and the Tsinghua–Tencent 100K (TT100K) traffic sign benchmark. Experiments show that our method has superior detection performance and is quicker than the general faster RCNN object detection framework on both datasets. It is competitive with state-of-the-art specialist traffic sign detectors on TT100K, but is an order of magnitude faster. To show generality, we tested it on the LISA dataset without tuning, and obtained an average precision in excess of 90%.

KW - CNN

KW - small object detection

KW - traffic light benchmark

KW - traffic light detection

U2 - 10.1007/s41095-018-0116-x

DO - 10.1007/s41095-018-0116-x

M3 - Article

VL - 4

SP - 253

EP - 266

JO - Computational Visual Media

T2 - Computational Visual Media

JF - Computational Visual Media

SN - 2096-0433

IS - 3

ER -