Framework for pedestrian walking behaviour recognition to minimize road accident

Pedestrian walking misbehaviour represents a severe problem to road safety. Therefore, pedestrian behaviour classification is a perfect solution in providing safety for both pedestrians and vehicles by exchanging movement information among entities via wireless communication. However, wireless...

Full description

Saved in:
Bibliographic Details
Main Author: Hashim Kareem, Zahraa
Format: Thesis
Language:English
English
English
Published: 2021
Subjects:
Online Access:http://eprints.uthm.edu.my/4128/1/24p%20ZAHRAA%20HASHIM%20KAREEM.pdf
http://eprints.uthm.edu.my/4128/2/ZAHRAA%20HASHIM%20KAREEM%20COPYRIGHT%20DECLARATION.pdf
http://eprints.uthm.edu.my/4128/3/ZAHRAA%20HASHIM%20KAREEM%20WATERMARK.pdf
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Pedestrian walking misbehaviour represents a severe problem to road safety. Therefore, pedestrian behaviour classification is a perfect solution in providing safety for both pedestrians and vehicles by exchanging movement information among entities via wireless communication. However, wireless communication has critical issues with network failure, and these issues significantly affect the communication system. Thus, the framework involved two modules for pedestrian walking behaviour classification in a vehicle-to-pedestrian (V2P) context is proposed. In the methodology, this study discloses five useful stages. Firstly, mobile phone users' irregular walking behaviour is investigated using a questionnaire to determine their options on mobile usage in the street. Secondly, four different testing scenarios are chosen to acquire pedestrian walking data using the gyroscope sensor, where the essential features were extracted and selected. Thirdly, the pedestrian's behaviour is recognized using grid optimizer in machine learning. Fourthly, four standard vectors for pedestrian walking behaviour are developed. Fifthly, the performance of the proposed classification methods is validated and evaluated against multiple scenarios and features. Two sets of real-time data are presented in this work. The first one is related to the questionnaire data, consisting of 262 respondent samples, while the second set has 263 samples of pedestrian walking signals. The results indicate the following: (1) From 262 samples, 66.80% and 48.10% of respondents use mobile phones for calling and chatting, respectively. (2) 263 samples of participants are obtained and analysed, and 90 features are extracted from each sample. (3) 100% classification accuracy are obtained for each class (normal walking, calling, chatting, and running) using the grid optimiser method in machine learning. (4) The precision of classification using Euclidean algorithm for normal walking and calling is 70%. In contrast, for chatting and running behaviour, the accuracy is 100% and 80%, respectively. This study's implication serves the safety system in the V2P context by programming the proposed framework as an application in smartphones for exchanging pedestrian information to the vehicles for avoiding accidents.