ANALYSING DEPTH DATA WITH SUPERVISED AND UNSUPERVISED LEARNING

This thesis investigates the application of machine learning algorithms for distance determination in the context of smart glasses technology. The primary objective is to enhance visual assistance capabilities by accurately predicting distance values, particularly focal length adjustments, based on sensor data. The study explores various machine learning methods, including k-means clustering, K-nearest neighbors (KNN) regression, Convolutional Neural Network (CNN) and DBSCAN clustering, to analyze their effectiveness in distance segmentation and prediction. The research begins with an exploration of adjustable lens technology and vision impairment concepts to establish foundational knowledge. Subsequently, machine learning algorithms are applied to extract meaningful insights from distance data collected by Time-of-Flight (ToF) sensors. Supervised and unsupervised learning techniques are employed for distance segmentation, enabling precise categorisation of depth measurements into meaningful clusters. Key findings from the study demonstrate the effectiveness of machine learning methods in accurately determining distance values, which is crucial for optimizing visual experiences and enhancing user interaction with smart glasses. Additionally, domain knowledge is integrated into the clustering process to inform k-selection and validate cluster consistency.