Abstract: To address the limitations of traditional crop phenotyping, including low efficiency, limited accuracy and destructive sampling, this paper systematically reviewed the applications of 3D point cloud processing technology in crop high-throughput phenotyping, and surveyed the extended application status of phenotype-driven smart agriculture. By summarizing crop 3D reconstruction technologies including LiDAR, multi-view stereo vision, and depth cameras, their robustness in complex agricultural scenarios was analyzed. Moreover, the evolution of crop point cloud processing algorithms which shifted from “conventional handcrafted feature engineering with machine-learning-based regression” towards “deep learning-based paradigms” was systematically reviewed, with emphasis placed on the advantages of deep learning models (such as PointNet++ and Transformer) in addressing non-rigid deformation, organ similarity and complex structure segmentation. This paper further elaborated on the current status of crop high-throughput phenotyping, covering digital-twin-based phenotypic trait temporal tracking and mainstream indoor and field highly integrated phenotyping platforms worldwide. On this basis, smart extended applications of phenotypic information were further reviewed, covering crop growth monitoring, early stress diagnosis, and the operational status of phenotypic-information-driven intelligent agricultural machinery in autonomous navigation, precision spraying, intelligent pruning, and automated harvesting. By systematically organizing existing technologies and methods, this paper aimed to reveal their advantages and limitations, provide references for innovative applications in crop phenotyping and intelligent agricultural equipment, and propose potential future research directions.