Seed shape is one of the most significant traits in plant identification and genetic blueprints. Geometric parameters of seeds can also play a pivotal role in the growth trajectories of high plants. Specifically, precise and efficient data acquisition can greatly contribute to digital seed evaluation and phenomics profiling in smart agriculture. It is required to accurately measure the geometric parameters of plant seeds. In this study, a high-throughput approach was introduced to measure the geometric parameters of seeds using a handheld scanner. Two solutions were also proposed for the challenges in conventional measurement, namely point cloud data missing and automatic screening of typical samples. Firstly, a point cloud completion was realized for the incomplete point cloud data during the batch scanning of diverse plant seeds. Ellipse fitting and smooth interpolation were employed to tailor the longitudinal profile contours of seeds, and then accurately bridge the gaps in the scanned point clouds. Data integrity and fidelity were then preserved for the downstream analysis. Secondly, an automatic selection of a typical sample was introduced to improve the efficiency using principal component analysis (PCA), in order to avoid the high labor intensity of traditional manual selection. The representative seed samples were then identified and extracted to dramatically enhance the data quality and consistency. According to the point cloud data and typical samples, 11 geometric parameters were measured automatically, including volume, surface area, length, width, height, as well as the perimeter and area of three principal component profiles. A high-precision handheld scanner, RigelScan Elite, was employed to capture the point cloud data. Eight types of plant seeds were also collected, such as broad beans, peanuts, kidney beans, soybeans, peas, black beans, red beans, and mung beans. 3,400 samples were captured to validate the measurement. There was the regular shape in the broad beans, peanuts, and kidney beans. While the irregular shape was found in the soybeans, peas, black beans, red beans, and mung beans. The 3D point cloud data of plant seeds was obtained with a resolution of 0.01 mm. Subsequently, algorithmic processing was performed on the color information for missing data completion, automatic screening, and measurement of representative samples. The results demonstrate that the point cloud completion was effectively utilized to successfully restore missing data with an average error of 0.017 mm under the various geometric regularities of seeds. The sample selection accurately identified the representative samples, with an average extraction error of 0.80%, indicating high precision. Furthermore, the automatic extraction of geometric parameters was integrated with an average measurement error below 0.41%. In conclusion, the high-throughput approach was presented to realize the batch, automatic, and high-precision measurement of seed geometric parameters. A reliable and efficient tool was also offered for the fundamental data acquisition and seed characterization in digital evaluation, phenomics, and smart agriculture.