Penerapan CNN Untuk Sistem Deteksi Personal Area Berbasis Citra Visual

Pritina, Odilla Kalya (2026) Penerapan CNN Untuk Sistem Deteksi Personal Area Berbasis Citra Visual. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5022221244-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (899kB) | Request a copy

Abstract

Konsep proxemics yang diperkenalkan oleh Edward T. Hall menjelaskan bahwa manusia mempertahankan batas ruang personal melalui empat zona interaksi, yaitu intimate, personal, social, dan public. Pemahaman terhadap zona-zona tersebut penting dalam pengembangan sistem cerdas yang mampu berinteraksi secara alami dengan manusia. Namun, implementasi konsep proxemics ke dalam sistem komputasional masih menghadapi tantangan, seperti keterbatasan metode visual konvensional dalam memahami konteks interaksi sosial, tingginya biaya penggunaan sensor khusus, serta belum tersedianya kerangka kerja yang mampu memodelkan prinsip proxemics secara efektif dan real-time. Penelitian ini bertujuan menerapkan Convolutional Neural Network (CNN) untuk sistem deteksi personal area berbasis citra visual menggunakan kamera RGB sebagai penangkap citra, mini PC sebagai unit pre-processing, dan metode YOLO11. Sistem memanfaatkan arsitektur YOLO11 yang terdiri atas backbone, neck, dan head untuk mendeteksi manusia serta menghasilkan 17 titik keypoint pose tubuh. Informasi pose dikombinasikan dengan data kedalaman dari kamera Intel RealSense melalui proses depth to color alignment, estimasi kedalaman, Kalman Filter, estimasi orientasi tubuh, dan perhitungan jarak antarindividu menggunakan Euclidean Distance untuk mengklasifikasikan zona proxemics secara real-time. Hasil pengujian menunjukkan akurasi klasifikasi proxemics sebesar 92,6% dan akurasi klasifikasi orientasi tubuh sebesar 73,2%. Pada tugas regresi estimasi kedalaman, model menghasilkan Mean Absolute Error (MAE) sebesar 86,799, nilai koefisien determinasi (R²) sebesar 0,604, serta deviasi sekitar 10–20% terhadap jarak aktual. Hasil tersebut menunjukkan bahwa sistem mampu memodelkan hubungan spasial antarindividu dengan baik dan menerjemahkan konsep proxemics ke dalam model komputasional yang dapat mendeteksi serta memvisualisasikan batas ruang personal antarindividu secara real-time.
======================================================================================================================================
The concept of proxemics, introduced by Edward T. Hall, explains that humans maintain personal space through four interaction zones: intimate, personal, social, and public. Understanding these zones is essential for developing intelligent systems capable of interacting naturally with humans. However, implementing the concept of proxemics in computational systems remains challenging due to the limitations of conventional vision-based methods in understanding social interaction contexts, the high cost of specialized sensors, and the lack of an effective real-time computational framework for modeling proxemics. This study aims to implement a Convolutional Neural Network (CNN) for a vision-based personal area detection system using an RGB camera for image acquisition, a mini PC for pre-processing, and the YOLO11 architecture. The proposed system utilizes the YOLO11 architecture, consisting of a backbone, neck, and head, to detect humans and estimate 17 body pose keypoints. Pose information is combined with depth data from an Intel RealSense camera through depth-to-color alignment, depth estimation, Kalman filtering, body orientation estimation, and Euclidean distance calculation to classify proxemics zones in real time. Experimental results show that the proposed model achieved a proxemics classification accuracy of 92.6% and a body orientation classification accuracy of 73.2%. For the depth estimation regression task, the model obtained a Mean Absolute Error (MAE) of 86.799, a coefficient of determination (R²) of 0.604, and an estimated distance deviation of approximately 10–20% from the actual distance. These results demonstrate that the proposed system effectively models spatial relationships between individuals and successfully translates the concept of proxemics into a computational framework capable of detecting and visualizing interpersonal personal space boundaries in real time.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	CNN, Personal Area, Proxemics, RGB, YOLO, Zona Interaksi.
Subjects:	T Technology > T Technology (General) > T59.7 Human-machine systems. T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques. Image analysis--Data processing. T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7882.P3 Pattern recognition systems
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20201-(S1) Undergraduate Thesis
Depositing User:	Odilla Kalya Pritina
Date Deposited:	21 Jul 2026 08:06
Last Modified:	21 Jul 2026 08:06
URI:	http://repository.its.ac.id/id/eprint/136062

Actions (login required)

View Item