Pratama, Mukhamad Nabila (2026) Kacamata Pintar Berbasis Raspberry Pi Dengan Deteksi Objek Dan Lubang Untuk Membantu Mobilitas Tunanetra. Other thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 5024221022-Undergraduate_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
5024221022-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (25MB) | Request a copy |
Abstract
Gangguan penglihatan secara signifkan membatasi kemampuan seseorang untuk bernavigasi dengan aman dalam lingkungan sehari-hari. Menurut Organisasi Kesehatan Dunia (WHO), lebih dari 2,2 miliar orang di seluruh dunia mengalami beberapa bentuk gangguan penglihatan, yang banyak di antaranya berujung pada kebutaan permanen. Penelitian ini menyajikan desain, implementasi, dan evaluasi sistem kacamata pintar wearable berbasis visi komputer (computer vision) untuk membantu penyandang disabilitas netra dalam mendeteksi rintangan dan lubang di sekitar mereka secara real-time. Sistem ini mengintegrasikan algoritme deteksi objek YOLO dengan platform edge computing Raspberry Pi 5 dan sebuah webcam USB, yang memberikan umpan balik audio kontekstual melalui earphone Bluetooth True Wireless Stereo (TWS). Sebuah dataset khusus berisi 897 gambar yang mencakup lima kelas bahaya—pejalan kaki, tiang, lubang jalan, penghalang jalan, dan pohon—dikumpulkan di lingkungan luar ruangan dan dianotasi menggunakan platform Roboflow. Penelitian ini mengimplementasikan model YOLOv8 nano sebagai algoritme utama, dengan YOLOv11 nano dilatih dan dievaluasi sebagai pembanding. Meskipun pada evaluasi awal YOLOv11 mencapai mAP@50 sebesar 0,816, presisi 0,855, dan F1-score 0,796, YOLOv8 dipilih untuk implementasi akhir karena efsiensi komputasinya. Dari empat konfgurasi inferensi yang diuji secara tolok ukur (YOLOv8 dan YOLOv11 dalam format ONNX Runtime serta format terkuantisasi TensorFlow Lite / TFLite INT8), throughput terbaik dicapai oleh YOLOv8 TFLite pada kecepatan 10,64 FPS dengan latensi rata-rata 92,4 ms. Kinerja ini berada jauh di bawah ambang batas persepsi pendengaran manusia sebesar 200–300 ms, menjadikannya paling ideal untuk navigasi waktu nyata meskipun YOLOv11 TFLite memiliki akurasi deteksi yang sedikit lebih tinggi (mAP@50 sebesar 0,769 dan presisi 0,767). Sistem ini menyampaikan peringatan audio terstruktur yang menginformasikan kelas objek, posisi arah jarum jam secara lateral, dan estimasi jarak, yang semuanya diproses secara lokal tanpa konektivitas internet. Integrasi fsik diwujudkan melalui bingkai kacamata PLA hasil cetak 3D, casing Raspberry Pi berpemasang sabuk dengan pendingin aktif, dan paket baterai Li-ion ganda 18650 dengan regulasi DC-DC. Hasil mengonfrmasi kelayakan perangkat wearable asistif berbasis edge-computing bagi pejalan kaki disabilitas netra dan menetapkan baseline yang kuat untuk akselerasi perangkat keras serta ekspansi dataset lebih lanjut.
========================================================================================================================================
Visual impairment signifcantly restricts an individual’s ability to navigate safely in everyday environments. According to the World Health Organization (WHO), more than 2.2 billion people globally experience some form of vision impairment, many of which result in permanent blindness. This research presents the design, implementation, and evaluation of a wearable smart glasses system based on computer vision to assist visually impaired individuals in detecting surrounding obstacles and potholes in real-time. The system integrates the YOLO object detection algorithm with a Raspberry Pi 5 edge computing platform and a USB webcam, delivering contextual audio feedback via Bluetooth True Wireless Stereo (TWS) earphones. A custom dataset of 897 images covering fve hazard classes—pedestrian, pole, pothole, roadblock, and tree—was collected in outdoor environments and annotated using the Roboflow platform. This research implements the YOLOv8 nano model as the primary algorithm, with YOLOv11 nano trained and evaluated as a comparison. Although in the initial evaluation YOLOv11 achieved a mAP@50 of 0.816, a precision of 0.855, and an F1-score of 0.796, YOLOv8 was selected for the fnal implementation due to its computational efciency. Out of the four inference confgurations benchmarked (YOLOv8 and YOLOv11 in both ONNX Runtime and INT8 quantized TensorFlow Lite / TFLite formats), the best throughput was achieved by YOLOv8 TFLite at 10.64 FPS with an average latency of 92.4 ms. This performance is well below the human auditory perception threshold of 200–300 ms, making it the most ideal for real-time navigation even though YOLOv11 TFLite has slightly higher detection accuracy (mAP@50 of 0.769 and precision of 0.767). The system delivers structured audio warnings communicating the object class, lateral clock-direction position, and estimated distance, all processed locally without internet connectivity. Physical integration was realized through a 3D-printed PLA glasses frame, a belt-mounted Raspberry Pi casing with active cooling, and a dual 18650 Li-ion battery pack with DC-DC regulation. The results confrm the feasibility of an edge-computing-based assistive wearable device for visually impaired pedestrians and establish a solid baseline for further hardware acceleration and dataset expansion
| Item Type: | Thesis (Other) |
|---|---|
| Uncontrolled Keywords: | Raspberry Pi 5, YOLO, OpenCV, Computer Vision, Kacamata Pintar, Deteksi Objek, Smart Glasses, Object Detection |
| Subjects: | T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T57.74 Linear programming T Technology > T Technology (General) > T58.5 Information technology. IT--Auditing 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 > TK5103.2 Wireless communication systems. Two way wireless communication |
| Divisions: | Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Computer Engineering > 90243-(S1) Undergraduate Thesis |
| Depositing User: | Mukhamad Nabila Pratama |
| Date Deposited: | 18 Jul 2026 05:15 |
| Last Modified: | 18 Jul 2026 05:15 |
| URI: | http://repository.its.ac.id/id/eprint/134936 |
Actions (login required)
 |
View Item |