Integrasi Citra Berwarna, NIR, dan Point Cloud Untuk Sistem Pengenalan Benda Rumah Tangga Dalam Pencahayaan yang Variatif

Rahmawati, Mawaddah Sekar (2026) Integrasi Citra Berwarna, NIR, dan Point Cloud Untuk Sistem Pengenalan Benda Rumah Tangga Dalam Pencahayaan yang Variatif. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022232004-Master_Thesis.pdf Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Domestic service robot merupakan salah satu terobosan teknologi untuk memudahkan kehidupan manusia. Robot jenis ini mencakup robot-robot yang digunakan di rumah tangga. Robot harus mampu mengenali objek (object recognition). Namun lingkungan rumah tangga memiliki kondisi pencahayaan yang berbeda-beda di mana object recognition yang hanya berbasis citra warna, kurang baik di kondisi tersebut. Untuk mengatasi keterbatasan tersebut, pengolahan data pada object recognition dengan memanfaatkan berbagai sensor seperti Near Infra Red (NIR) dan depth (D) menjadi salah satu solusi yang potensial. Data NIR tahan terhadap perubahan cahaya sedangkan data depth diubah menjadi data Point Cloud yang merupakan format populer untuk mewakili bentuk 3D objek. Integrasi dirancang untuk mengatasi perubahan pencahayaan, dengan memanfaatkan kombinasi fitur dari data RGB, NIR dan point cloud. Penelitian ini bertujuan untuk mengintegrasikan fitur semantik dari data RGB, dan fitur geometric dari data NIR dan point cloud menggunakan metode conditional late fusion. Serta untuk mengetahui akurasi sistem integrasi RGB-NIR dan point cloud pada object recognition benda rumah tangga dalam kondisi pencahayaan yang variatif. Dalam penelitian ini objek yang akan dikenali berupa barang-barang rumah tangga. Berdasarkan pengujian model integrasi citra RGB, NIR dan point cloud didapatkan akurasi sebesar 93,51% dan F1-score sebesar 93,15%. Pada pengujian model dalam kondisi pencahayaan gelap, ambient light, lampu jingga, dan lampu ungu, didapatkan akurasi sebesar 90,10%, 88,54%, 88,54%, dan 90,62%, secara berurutan. Hasil ini menunjukkan model mampu mengenali objek rumah tangga dengan baik pada berbagai kondisi pencahayaan yang diuji.
=========================================================================================================================
Domestic service robots represent a technological breakthrough designed to facilitate human life. This category encompasses robots utilized within household environments. These robots must be capable of object recognition. However, household environments possess varying lighting conditions, where object recognition based solely on colour imagery (RGB) performs poorly. To overcome this limitation, data processing for object recognition utilizing various sensors, such as Near Infrared (NIR) and depth (D), offers a potential solution. NIR data is robust against lighting changes, while depth data is converted into Point Cloud data, which is a popular format for representing the 3D shape of objects. This integration is designed to address lighting variations by leveraging a combination of features from RGB, NIR, and point cloud data. This study aims to integrate semantic features from RGB data with geometric features from NIR and point cloud data using a conditional late fusion method. Furthermore, it aims to determine the accuracy of the RGB-NIR and point cloud integration system for household object recognition under varying lighting conditions. In this study, the objects targeted for recognition consist of household items. Based on the testing of the RGB, NIR, and point cloud integration model, an accuracy of 93,51 and an F1-score of 93,15 were obtained. During model testing under dark conditions, ambient light, orange lighting, and purple lighting, accuracies of 90,10%, 88,54%, 88,54%, and 90,62% were obtained, respectively. These results demonstrate that the model is capable of effectively recognizing household objects across various lighting conditions during the testing process.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	efficientNet, object recognition, ResNet, pointNet, citra berwarna, NIR, point cloud, efficientNet, object recognition, ResNet, pointNet, RGB, NIR, point cloud
Subjects:	T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T57.74 Linear programming
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20101-(S2) Master Thesis
Depositing User:	Mawaddah Sekar Rahmawati
Date Deposited:	21 Jan 2026 07:44
Last Modified:	21 Jan 2026 07:44
URI:	http://repository.its.ac.id/id/eprint/129975

Actions (login required)

View Item