Design and Validation of Autonomous Perception Pipelines: Knowledge Transfer from Micromobility to Macromobility

In recent years, research on Advanced Driver Assistance Systems (ADAS) has gained significant momentum, due to the growing need for technologies that enhance safety, comfort and efficiency in transportation. ADAS are designed to support drivers in both Micromobility and Macromobility devices by monitoring the surroundings with sensors, cameras, RADAR and LiDAR. These systems provide real-time data to assist the driver navigate safely by detecting potential hazards, such as pedestrians or other vehicles. Through early warnings or automatic interventions, ADAS help to prevent collisions while also supporting drivers to maintain safe distances, adjust the driving speed to traffic and assist with lane-keeping. To ensure the effective development of these advanced systems, a scalable and safe testing environment is essential. This allows for controlled and progressive testing of ADAS, minimizing risks and ensuring accurate validation. Using Micromobility devices instead of Macromobility vehicles reduces the potential for serious accidents during early testing phases. These smaller and lightweight vehicles operate at lower speeds, making it safer to identify and correct system errors, as well as accelerating the implementation of ADAS across various modes of transportation. This thesis primarily focuses on Micromobility devices, leveraging their unique advantages to develop innovative ADAS solutions. These vehicles provide a cost-effective and flexible platform for testing and experimentation, which can later be transferred to Macromobility systems. By taking advantage of their lower cost, ease of implementation and simpler architecture, Micromobility vehicles serve as ideal testbed for developing and refining ADAS that can be scaled up to more complex and larger transportation systems. A key aspect of the research is the development of an Image Processing pipeline for Object Detection, focusing on the optimization and real-time computation of the input data. The pipeline validation was performed using a Micromobility platform developed in collaboration with the company Teoresi S.p.A, on which two perception sensors were installed: a stereocamera and a 4D RADAR. This thesis emphasizes the pivotal role of Micromobility devices as a bridge for ADAS development, leveraging their cost-effectiveness and flexibility to generate insights and innovations that can be applied to more complex transportation systems.