AI-powered safe bicycling routes for sustainable urban mobility in Malta

Abstract

Urban mobility and cyclist safety represent critical challenges in modern transportation infrastructure, particularly in compact urban environments like Malta. This research project explores an AI-based system for generating safe cycling routes by integrating computer vision, machine learning, and geospatial analysis techniques. The proposed solution leverages a dataset of Strava cycling segments supplemented with additional generated routes, employing a methodology that combines visual safety classification, elevation data, and overall distance. A convolutional neural network (CNN) is developed to classify route safety levels, utilising multiperspective Google Street View images to assess potential hazards and route characteristics. The CNN model achieved 82.3% classification accuracy with particularly strong performance at identifying extreme safety conditions (89.1% for safest routes, 86.7% for most dangerous). User evaluation with 8 participants demonstrated 76.2% agreement between AI safety assessments and human perception. Comparative analysis revealed that users preferred AI-generated routes over conventional navigation systems in 70% of cases, with expert validation showing 73.3% correlation between algorithmic and professional cyclist assessments. By integrating computational techniques with detailed geographical analysis, the project demonstrates a holistic approach to route recommendation that prioritises cyclist safety and promotes sustainable urban mobility. The results validate the feasibility of AI-powered cycling safety assessment, offering a data-driven methodology for identifying and recommending safe cycling routes that can be adapted to various urban environments.