High-throughput phenotyping of plant three-dimensional (3D) architecture is critical for determining plant phenotypic characteristics. The acquisition of 3D architecture of plant phenotypic traits based on machine vision has been widely applied in greenhouse research. Growth process of the plants can be dynamically monitored. However, the application of machine vision method in the field is less due to the complex environment. Machine vision method was used to obtain multi-view image sequences for field growth maize and soybean at different growth stages. Then 3D architecture of individual plants and its populations were reconstructed. The accuracy of calculated individual blade length and maximum width was evaluated according to the measured data. The results showed that there was a good agreement between measured and calculated blade length and blade maximum width with R2 which were both more than 0.97. Then the dynamic changes of plant height, crown surface and organ growth were extracted based on reconstructed 3D architecture automatically. The results can provide a method for high throughput phenotypic analysis related to genotypes and help to evaluate the plant architecture and canopy radiation.