Perancangan Prototipe Asistif Untuk Navigasi Dalam Ruangan Menuju Pintu Keluar Bagi Penyandang Tunanetra

Hutahaean, Nathaniel Parsaoran (2026) Perancangan Prototipe Asistif Untuk Navigasi Dalam Ruangan Menuju Pintu Keluar Bagi Penyandang Tunanetra. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5022221095-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (31MB) | Request a copy

Abstract

Penelitian ini merancang dan menguji prototipe asistif navigasi dalam ruangan untuk membantu penyandang tunanetra mencapai pintu keluar secara mandiri. Prototipe menggunakan kamera stereo OAK-D dengan komputasi terdistribusi antara Raspberry Pi 5 sebagai node sensor dan aktuator serta laptop sebagai pemroses utama. Deteksi pintu dilakukan menggunakan model YOLO26, pelacakan pintu dikelola oleh Goal-Lock Finite State Machine (FSM), posisi pengguna diestimasi melalui Pedestrian Dead Reckoning (PDR) berbasis IMU, dan navigasi berbasis celah digunakan untuk memilih koridor kosong di depan pengguna. Ketika pintu tidak terlihat, Vision-Language Model (VLM) berperan sebagai *landmark advisor* yang memilih objek di lantai sebagai titik pandu perantara setelah melalui validasi kedalaman, posisi, dan arah. Instruksi disampaikan dalam bentuk perintah gerak diskret berbahasa Indonesia melalui Text-to-Speech (TTS), dilengkapi dengan umpan balik buzzer dan tombol konfirmasi fisik sebagai mekanisme *turn-taking*. Pengujian di Laboratorium AJ-303 yang melibatkan 18 percobaan pada tiga posisi awal dan dua kondisi VLM menunjukkan bahwa 14 dari 18 percobaan berhasil mencapai pintu. Kondisi VLM aktif berhasil pada 8 dari 9 percobaan (88,9%), sedangkan kondisi VLM nonaktif berhasil pada 6 dari 9 percobaan (66,7%). Nilai rata-rata Intersection over Union (IoU) *bounding box* landmark sebesar 0,426 dengan latensi respons rata-rata 3,67 detik. Satu percobaan pada kondisi VLM aktif gagal karena landmark mengarahkan pengguna ke jalur di luar rute menuju pintu, yang menunjukkan bahwa keluaran VLM masih memerlukan verifikasi berbasis aturan sebelum digunakan sebagai tujuan navigasi.
=================================================================================================================================
This research designs and evaluates an indoor navigation assistive prototype to help visually impaired individuals independently reach exit doors. The prototype employs an OAK-D stereo camera with distributed computing between a Raspberry Pi 5 serving as the sensor and actuator node and a laptop functioning as the main processing unit. Door detection is performed using a custom-trained YOLO26 model, door tracking is managed by a Goal-Lock Finite State Machine (FSM), user position is estimated through IMU-based Pedestrian Dead Reckoning (PDR), and gap-based navigation is used to identify open corridors ahead of the user. When the door is not visible, a Vision-Language Model (VLM) acts as a landmark advisor by selecting floor-level objects as intermediate waypoints after validating their depth, position, and direction. Navigation instructions are delivered as discrete movement commands in Indonesian through a Text-to-Speech (TTS) system, complemented by buzzer feedback and a physical confirmation button implementing a turn-taking mechanism. Experiments conducted in Laboratory AJ-303 with 18 trials across three starting positions and two VLM conditions showed that 14 of the 18 trials successfully reached the exit door. The active VLM condition achieved success in 8 of 9 trials (88.9%), whereas the inactive VLM condition succeeded in 6 of 9 trials (66.7%). The average landmark bounding box Intersection over Union (IoU) was 0.426, with a mean response latency of 3.67 seconds. One trial under the active VLM condition failed because the selected landmark directed the user onto a path outside the route to the exit door, indicating that VLM outputs still require rule-based verification before being adopted as navigation goals.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Navigasi Dalam Ruangan, Tunanetra, OAK-D, YOLO26, Goal-Lock FSM, PDR, Vision-Language Model, Landmark Advisor ndoor Navigation, Visually Impaired, OAK-D, YOLO26, Goal-Lock FSM, PDR, Vision-Language Model, Landmark Advisor
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence 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.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20201-(S1) Undergraduate Thesis
Depositing User:	Nathaniel Parsaoran Hutahaean
Date Deposited:	23 Jul 2026 01:16
Last Modified:	23 Jul 2026 01:16
URI:	http://repository.its.ac.id/id/eprint/135994

Actions (login required)

View Item