Simulation Of Object Detection And Classification Using Machine Learning

No Thumbnail Available

Date

2021-12

Journal Title

Journal ISSN

Volume Title

Publisher

Department of Electrical and Electronic Engineering

Abstract

Simulation Of Object Detection And Classification Using Machine Learning Risk factors of road accidents are increasing around the modern world. It is also applicable to Bangladesh. To reduce road accidents we proposed a system that detects objects on the road. We can implement this system using the available pre-trained model. Unfortunately, most available pre-trained models cannot detect Bangladeshi vehicles shapes and patterns. Thus, Our approach is to make a custom dataset and model from scratch. Based on this dataset and model irregular-shaped vehicles that are available in Bangladesh can be detected. To achieve our proposed method, we have collected Bangladeshi Vehicles Dataset containing 248 images. Based on this dataset a custom model has been built. We tested another 82 images which are distinct from the dataset to verify the accuracy of the custom model. We got a roughly 65 percent accuracy rate to detect these native vehicles varying the range from 91 - 37 percent. Thus, this system can be implemented as a road collision avoidance system in developing countries like Bangladesh.

Description

submitted by S. M. Irfan, bearing Matric ID. ET163077 and Muhammad Iftekhar Chowdhury, bearing Matric ID. ET163083 of session Spring 2021

Keywords

Citation