Bangladeshi Local Vehicle Recognition with A Comprehensive Dataset using Transfer Learning Techniques

Vehicle detection, localisation, and classification are essential for developing an Intelligent Transport System (ITS). Bangladesh is a developing nation, and the number of vehicles here is increasing significantly each year. Therefore, the development of an intelligent transportation system has bec...

Full description

Saved in:
Bibliographic Details
Published in2022 4th International Conference on Sustainable Technologies for Industry 4.0 (STI) pp. 1 - 6
Main Authors Rahman, Md Sazedur, Hassan, Md Zahim, Hossain, Syed Nahin, Masrur, Noor, Rabbi, Jakaria
Format Conference Proceeding
LanguageEnglish
Published IEEE 17.12.2022
Subjects
Bangladeshi vehicle dataset
deep learning
machine learning
object detection
pre-trained model
Road accidents
Roads
Surveillance
Traffic control
Transfer learning
Urban areas
Vehicle detection
Online AccessGet full text

Cover

More Information
Summary:Vehicle detection, localisation, and classification are essential for developing an Intelligent Transport System (ITS). Bangladesh is a developing nation, and the number of vehicles here is increasing significantly each year. Therefore, the development of an intelligent transportation system has become crucial. In order to assess the models' performance and the dataset's suitability for the task, we developed a demanding dataset of Bangladeshi vehicles and trained a variety of pre-trained models using it. The dataset contains 12,413 images of 9 most common local vehicle types of Bangladesh.We carefully examined the performance of the model on the image database. We proposed a model that provides the most accurate results for the localization, and categorization tasks of Bangladeshi vehicles. The study also indicates that the newly introduced dataset is more challenging and offers more similar scenarios that resembles real life situations than pre-existing datasets.
DOI:10.1109/STI56238.2022.10103325