Firearm-related action recognition and object detection dataset for video surveillance systems

The proposed dataset is comprised of 398 videos, each featuring an individual engaged in specific video surveillance actions. The ground truth for this dataset was expertly curated and is presented in JSON format (standard COCO), offering vital information about the dataset, video frames, and annotations, including precise bounding boxes outlining detected objects. The dataset encompasses three distinct categories for object detection: "Handgun", "Machine_Gun", and "No_Gun", dependent on the video's content. This dataset serves as a resource for research in firearm-related action recognition, firearm detection, security, and surveillance applications, enabling researchers and practitioners to develop and evaluate machine learning models for the detection of handguns and rifles across various scenarios. The meticulous ground truth annotations facilitate precise model evaluation and performance analysis, making this dataset an asset in the field of computer vision and public safety.