Robbery and Coercion Action Recognition Dataset

The current security industry faces increasingly complex threats, especially in retail and public places, where violent behaviors of customers or personnel pose serious challenges to the safety of merchants and customers. However, existing monitoring systems often fail to effectively identify and respond to these behaviors in real-time, leading to increased losses and security risks. This dataset aims to help develop more accurate and efficient action detection algorithms by providing high-quality video data to enhance the intelligence level of security systems. Data collection is conducted using high-resolution cameras in real-world scenarios, covering various lighting and environmental conditions. Through multiple rounds of annotation and expert review, the accuracy and consistency of action annotation are ensured. Data storage is in MP4 format with each video file corresponding to an annotation file for easy subsequent processing and analysis. The core advantage of this dataset is its high annotation accuracy over 95%, with strict consistency checks to ensure data integrity and reliability. Additionally, new data augmentation techniques are employed to enhance the model's generalization ability, potentially improving performance metrics in action recognition tasks by at least 15%.