Abstract

A pedestrian is classified as a Vulnerable Road User (VRU) due during walking or crossing in the road pedestrian is not protected or shielded. This caused pedestrians to have the most potential risk than other road users, such as motorcycle drivers or car drivers. To support Autonomous vehicles (AV) toward a higher level of independence, AV needs to recognize pedestrian and behavior related to it. Our proposed method utilizes a deep learning approach to predict pedestrian behavior using eight pedestrian input features with three frame values: five frames, ten frames, and 15 frames. Each number of frames is consists of four models, with one hidden layer, two hidden layers, three hidden layers, and four hidden layers. To improve the deep learning model, we conduct hyperparameter tuning, including hidden layer parameters and a number of frame numbers. Our model can predict pedestrians to cross or not cross using eight input features, with the best model using a number of frames values ten combined with three hidden layers. The resulting model prediction accuracy is 94.77%, and the model prediction loss is 0.18%.

Original languageEnglish
Title of host publicationProceedings - 2021 International Seminar on Intelligent Technology and Its Application
Subtitle of host publicationIntelligent Systems for the New Normal Era, ISITIA 2021
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages420-425
Number of pages6
ISBN (Electronic)9781665428477
DOIs
Publication statusPublished - 21 Jul 2021
Event2021 International Seminar on Intelligent Technology and Its Application, ISITIA 2021 - Virtual, Online
Duration: 21 Jul 202122 Jul 2021

Publication series

NameProceedings - 2021 International Seminar on Intelligent Technology and Its Application: Intelligent Systems for the New Normal Era, ISITIA 2021

Conference

Conference2021 International Seminar on Intelligent Technology and Its Application, ISITIA 2021
CityVirtual, Online
Period21/07/2122/07/21

Keywords

  • Pedestrian behavior prediction
  • deep learning
  • number of frame

Fingerprint

Dive into the research topics of 'Pedestrian crossing decision prediction based on behavioral feature using deep learning'. Together they form a unique fingerprint.

Cite this