Recognizable Point-cloud: An end-to-end point-cloud processing, visualization, and user-interaction design

Point clouds have been widely used in various industries, such as autonomous driving, monitoring, and surveying. However, the visualization methods developed for these fields are limited compared to the rapidly evolving techniques for point cloud processing. Most of these visualization approaches are primarily focused on technical considerations, such as memory usage, computation latency, and visualization accuracy, rather than the user experience. The aim of this thesis is to design a new framework for point cloud visualization that incorporates state-of-the-art techniques and to evaluate its effectiveness through user recognition testing. The work involves the following steps: (1) conducting a literature review and comparing different baseline models for learning-based point cloud techniques, including detection and upsampling; (2) constructing a pipeline to train models and generate multiple point cloud visualization options; (3) developing a user interface with a set of questions on object, semantic, and spatial recognition; and (4) conducting user research to assess the visualization options. This study provides a comprehensive review of point cloud upsampling and detection methods, and presents a novel framework for enhancing point cloud visualization through the incorporation of advanced techniques. It also contributes to user-centered research on point clouds by introducing various forms of recognition as key factors.