Optimized Obstacle Detection for the Visually Impaired Using a Machine Learning-Based Modified Map Function for Ultrasonic Sensors

The Lancet Global Health Commission on Global Eye Health reported in 2020 that the number of people with visual impairment is rising, with a total estimate of 43 million being blind worldwide. Numerous studies were conducted, aiming to develop an assistive technology using ultrasonic waves to detect obstacles. However, a limited study that develops an assistive technology with the integration of ultrasonic sensors for obstacle detection and optimizes its system through machine learning (ML). In response, this study proposed an adaptive system that accurately detect and navigate around obstacles by integrating ML-based modified map function methods. Specifically, design a wearable obstacle detection technology using integrated microcontrollers, ultrasonic, audio, and vibration; develop the system using a ML algorithm; and optimize the accuracy using the improved system of wearable obstacle detection technology. Among the 10 ML regression algorithms, the ridge regression (L2) was the best ML suited for proposed modified map function $(MSE=9.36,\ R^{2}=1)$, whereas its generated regression equation $(\mathrm{y}=92.23+1.04x)$ is used to optimize its obstacle detection performance. The proposed ML-based modified map function system improved the device's obstacle detection performance by 99% compared to the traditional map function's 97. 77% accuracy. As a result, the performance of ultrasonic sensors in detecting obstacles was maximizes utilizing a ML-based modified map function method - thus making safer utilization for visually impaired obstacle detection.