In this study, two types of worker posture recognition systems were developed to recognize 18 fundamental worker posture types. The two systems used worker silhouette features and the eight color marker posture types to express the targeted worker posture types. In the first system, a posture lookup table was designed to correct the misrecognized postures. In the second system, a random forests algorithm of a machine learning algorithm was used to train the decision trees to recognize targeted postures. Additionally, two worker assistance systems, which were a pea field task monitoring system and a container carrying posture recognition system, were developed to demonstrate an applicability of the two posture recognition systems for an agricultural field. In our experiments, the total recognition rates of the two worker posture recognition systems were respectively around 90%. Moreover, the pea field task monitoring system could robustly track the workflow of watering and seeding for producing peas. The container carrying posture recognition system could robustly detect the carrying postures. The experimental results demonstrated the high performance and applicability of the two worker posture recognition systems for developing an autonomous agricultural robot.