DESIGN & IMPLEMENTATION OF DATA CONTROL SUBSYSTEM AND MACHINE LEARNING ALGORITHM ON VISUALLY IMPAIRED ADAPTIVE WHITE CANE

• Main Author: Fauzan, Muhammad
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/87715
• Institution: Institut Teknologi Bandung

White canes are widely used devices to assist blind people in mobility. However, these canes have limitations in terms of sensing range, information systems, and manufacturing quality. To overcome these challenges, an adaptive white cane was developed that is designed to detect obstacles on demand and provide real-time feedback to users. The device integrates a Raspberry Pi proximity sensor and camera to detect obstacles, paired with a Bluetooth earphone and vibration module to provide tactile and audio feedback. To ensure the adaptive cane can be used smoothly, the threading concept is carried out in three modes: distance detection mode, obstacle detection mode, and instruction mode. By utilizing machine learning algorithms and the Tensorflow library, among the tested models, the SSD-MobileNet-V2-FPNLite320 model achieves the highest obstacle detection accuracy with scores of 86.89%, 77.2%, 84.52%, and 77.09% for the trench, ramp, up stairs, and down stairs classes, respectively