Real-time Monocular Collision Alert System for Enhancing Safety in Micromobility Transportation

In the contemporary era, machine learning has emerged as a transformative force, reshaping industries and redefining the paradigms of safety and efficiency. Particularly in urban transportation, where the adoption of micro-mobility solutions like micro-electric vehicles has surged, the integration of machine learning techniques offers promising avenues to address emerging challenges. This thesis presents a pioneering approach to harnessing the power of machine learning for the design and deployment of a real-time monocular collision alert system tailored to enhance safety standards in micro-mobility transportation. Advanced object detection and tracking algorithms, underpinned by machine learning principles, form the core of this system, enabling it to swiftly and accurately identify potential collision threats in dynamic urban settings. A standout feature of this research is the innovative alert policy, which evaluates the trajectories and relative speeds of tracked objects, determining their collision risk potential. When a threat is imminent, the system dispatches timely alerts, granting riders crucial moments to react, a testament to the practical application of machine learning in ensuring safety. Extensive real-world evaluations underscore the system's efficacy. Beyond its precision in object detection and tracking, its alert mechanism stands out for its timeliness and accuracy. Notably, this system achieves its objectives without the reliance on high-cost sensors typical in modern vehicles, positioning it as a cost-effective and scalable solution. As the world leans more towards sustainable and efficient transportation solutions, the integration of machine learning not only enhances safety but also paves the way for a more responsive and adaptive urban transportation landscape.