<p>This paper presents a comprehensive, system-level survey of point cloud processing, motivated by the growing importance of point clouds as a core 3D representation in applications such as autonomous driving, robotics, medicine, and urban planning. The survey jointly reviews the principal high-level tasks of classification, segmentation, object detection, tracking, and compression, covering both traditional approaches and recent deep learning advances. Particular emphasis is placed on transformer-based architectures and learning-driven compression techniques, which have recently influenced the design of modern point cloud pipelines across multiple tasks. To reflect real-world 3D perception systems, the literature is organized using a task-oriented taxonomy that distinguishes between vision-level analysis tasks, which focus on semantic and geometric understanding, and signal-level operations such as compression, which directly impact data fidelity, computational efficiency, and downstream learning performance. The paper further clarifies the distinction between point cloud processing as a complete computational pipeline and point cloud analysis as semantic inference. By integrating learning-based compression into the broader processing framework and consolidating datasets, performance comparisons, and research timelines, this survey enables cross-task insights into design trends, maturity levels, and open challenges that are fragmented in existing literature.</p>