Recently, three-dimensional (3D) sensing has gained a lot of attention in plant phenotyping because it can accurately represent the 3D nature of plant architecture. Among all available 3D image sensor technologies, stereo vision offers a viable solution due to its high spatial resolution and wide selections of camera modules. However, the performance of in-field stereo imaging for plant phenotyping is adversely affected by textureless regions, occlusions, and outdoor lighting variations. In the field, shadows and sunlight can lead to under- or over-exposed images. In addition, movements of plant canopy caused by wind can result in motion blurs. These constraints have reduced the utility of stereo imaging to plant phenotyping in the field. In this research, a portable stereo imaging module was developed for high-throughput in-field phenotyping. This stereo module is capable of taking images at high frame rates with strobe lighting and producing high-quality images in outdoors conditions. We used a custom-built ground vehicle to deploy the stereo module to collect images of tightly spaced rows of sorghum and maize plants. An automated processing pipeline was developed to quantify stem diameter of the imaged plants, which is an important trait for stalk strength and biomass potential evaluation. The pipeline used Mask-RCNN for detecting stalk contours and Semi-Global Block Matching for generating disparity maps. Metric measurements were obtained based on the disparity maps and found to be highly correlated with the in-filed manual measurements. The results establish the feasibility of using the developed stereo module for acquiring high-quality stereo image pairs for semantic segmentation and precise plant trait measurements in 3D space. Furthermore, the proposed method could be a promising tool for high-throughput phenotyping for attaining other plant morphological traits (e.g. leaf angle, panicle size).