Children's Action Recognition Video Dataset in Game Room

In the fields of intelligent security and human-computer interaction, recognizing children's actions and behaviors is an important technical issue. However, existing solutions have limitations in recognition accuracy and adaptability, often struggling to handle complex action diversity and occlusion issues. This dataset aims to address this challenge by providing high-quality video data to support researchers in developing more accurate action recognition models. The dataset is primarily collected in various environments such as homes, kindergartens, and game rooms, using high-definition camera equipment to ensure coverage of a variety of action types. Multiple rounds of annotation and consistency checks are conducted to ensure data quality, and the annotation team is composed of experienced action research experts with more than 10 members. Preprocessing techniques such as video denoising, inter-frame alignment, and action clipping are employed to guarantee data consistency and usability. The data is stored in MP4 format and organized by time and action type.