Camera Obstruction Action Recognition Dataset

In the current security industry, as surveillance technology becomes more widespread, actions that obstruct cameras are increasingly frequent, posing a severe challenge to security monitoring. Traditional surveillance systems often fail to effectively identify such actions, leading to security risks. This dataset aims to provide a high-quality video dataset to support in-depth research and model training on obstruction actions, thereby enhancing the intelligence level of surveillance systems. Data collection was conducted using high-definition cameras installed in different environments, capturing various video samples of obstruction actions. To ensure data quality, a combination of multiple rounds of annotation and expert review was used for data validation. The data is stored in MP4 format for ease of subsequent processing and analysis.