Strengthening the safety management in confined space operations is an important foundation for preventing and reducing production safety accidents. In the grain soil, which are large enclosed spaces insufficient lighting and restricted air circulation create significant safety risks during the operation. Utilizing in-silo surveillance video to detect and analyze the behavior of workers is a crucial technical measure to ensure safe operations. This paper summarizes the methods for establishing and preprocessing datasets for video-based abnormal behavior detection in grain storage operations, explains the progress of machine learning and deep learning technologies in this field, including technological innovation and practical applications in areas such as abnormal behavior detection and real-time early warning. In addition, the paper reviews the research findings and existing problems in this field, such as incomplete datasets, insufficient model accuracy, etc., and provides an outlook on future research directions.