A Stereo-Integrated Novel View Synthesis Pipeline for the Enhancement of Road Surface Reconstruction Dataset

This proposal outlines a novel view synthesis pipeline designed for road reconstruction in autonomous driving scenarios that leverages virtual camera technology to synthesise images from unvisited camera poses, thereby enhancing and expanding current datasets. It consists of three main steps: data acquisition, data preprocessing and fusion, and then importantly interacting with new 3D view synthesis with geometric priors. The modular design allows each component to be independently optimised and upgraded, ensuring flexibility and adaptability to various datasets and task requirements. The proposed approach aims to improve the robustness, realism, and photometric consistency of novel view synthesis, effectively handling dynamic scenes and varying lighting conditions. Additionally, this research plans to open-source a low-cost stereo camera hardware solution with the included software implementation.