Sistem Kontrol Kursi Roda dengan Navigasi Paving Taktil dan Obstacle Avoidance Berbasis Computer Vision

Ashanti, Febrian Syafina Khaira (2026) Sistem Kontrol Kursi Roda dengan Navigasi Paving Taktil dan Obstacle Avoidance Berbasis Computer Vision. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5023221037-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (2MB) | Request a copy

Abstract

Navigasi mandiri pada kursi roda menjadi tantangan tersendiri bagi penyandang tunanetra akibat keterbatasan dalam mengikuti jalur paving taktil serta mengenali rintangan di lingkungan sekitar. Penelitian ini mengembangkan sistem kursi roda semi-otonom yang mengintegrasikan computer vision untuk navigasi berbasis paving taktil dengan mekanisme obstacle avoidance. Deteksi paving taktil dilakukan menggunakan model deep learning YOLOv8n yang mengenali enam kelas pola navigasi (lurus, belok kanan, belok kiri, pertigaan, perempatan, dan tanda berhenti), dijalankan secara real-time pada perangkat Jetson Xavier NX. Sensor ultrasonik HC-SR04 digunakan sebagai lapisan keselamatan tersendiri yang dikendalikan mikrokontroler STM32F1 untuk mendeteksi halangan dan menghentikan kursi roda, sementara tombol dengan penanda taktil berfungsi sebagai mekanisme verifikasi arah pada persimpangan. Pengujian melibatkan empat subjek berpenglihatan normal yang dikondisikan menyerupai dua kondisi gangguan penglihatan, yaitu full blind (tanpa penglihatan) dan partiall blind (penutup dengan keterbatasan lapang pandang), masing-masing dua subjek. Hasil pengujian menunjukkan bahwa model deteksi mencapai nilai mAP@0.5 sebesar 0.969 dengan akurasi deteksi real-time sebesar 91.4% pada perangkat Jetson Xavier NX. Selain itu, kelima sensor ultrasonik menunjukkan nilai Mean Absolute Error (MAE) kurang dari 1 cm. Pada pengujian navigasi keseluruhan, sistem mencapai tingkat keberhasilan 84.5% hingga 100% di seluruh subjek, dengan beban kerja pengguna yang tergolong rendah hingga sedang berdasarkan pengukuran NASA-TLX. Tingkat keberhasilan terendah ditemukan pada manuver belok, sehingga aspek tersebut masih memerlukan pengembangan lebih lanjut. Sistem yang dikembangkan menunjukkan potensi dalam meningkatkan kemandirian dan keselamatan mobilitas penyandang tunanetra.
======================================================================================================================================
Independent wheelchair navigation poses a particular challenge for visually impaired individuals due to difficulties in following tactile paving paths and recognizing obstacles in the surrounding environment. This research develops a semi-autonomous wheelchair system that integrates computer vision for tactilepaving-based navigation with an obstacle avoidance mechanism. Tactile paving detection is performed using the YOLOv8n deep learning model, which recognizes six navigation pattern classes (straight, turn right, turn left, T-junction, intersection, and stop sign), running in real time on a Jetson Xavier NX device. An HC-SR04 ultrasonic sensor array serves as an independent safety layer controlled by an STM32F1 microcontroller to detect obstacles and stop the wheelchair, while tactilemarked buttons function as a direction-verification mechanism at junctions. Testing involved four normally-sighted subjects conditioned to resemble two visual impairment conditions, namely fully blind (no vision) and partially blind (a restricted field of vision), with two subjects each. The results show that the detection model achieved an mAP@0.5 of 0.969 with a real-time detection accuracy of 91.4% on the Jetson Xavier NX, while the five ultrasonic sensors exhibited a Mean Absolute Error (MAE) of less than 1 cm. In the overall navigation testing, the system achieved success rates ranging from 84.5% to 100% across all subjects, with a user workload categorized as low to moderate based on NASATLX measurements. The lowest success rate was found in turning maneuvers, indicating that this aspect still requires further development. The developed system demonstrates potential for improving the independence and mobility safety of visually impaired users

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Computer Vision, Navigasi Paving Taktil, Obstacle Avoidance, Kursi Roda Semi-Otonom, YOLOv8n, Computer Vision, Tactile Paving Navigation, Obstacle Avoidance, Semi-Autonomous Wheelchair, YOLOv8n
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7882.P3 Pattern recognition systems
Divisions:	Faculty of Electrical Technology > Biomedical Engineering > 11410-(S1) Undergraduate Thesis
Depositing User:	Febrian Syafina Khaira Ashanti
Date Deposited:	01 Aug 2026 02:43
Last Modified:	01 Aug 2026 02:43
URI:	http://repository.its.ac.id/id/eprint/138671

Actions (login required)

View Item