Emanuel Trabes

Visual SLAM is crucial for technologies needing intricate scene understanding, such as self-driving cars and assistive devices for the visually impaired. However, VSLAM systems face challenges like mapping in complex conditions (e.g., heavy rain) and handling reflective and refractive surfaces, which break the “brightness constancy” assumption. The proposal aims to develop a more robust Dense VSLAM (DVSLAM) system capable of functioning in environments with reflective and refractive surfaces. The new system will be rooted in advancements in differentiable rendering. It must operate in real-time and be efficient in terms of processing power and energy consumption, given the often limited resources available. The research specifically targets applications in indoor navigation for the visually impaired and autonomous driving in complex weather conditions. The goal is to improve the state of the art in scene understanding technologies