Abstract: A computer-implemented method is provided for action localization. The method includes converting one or more video frames into person keypoints and object keypoints. The method further includes embedding position, timestamp, instance, and type information with the person keypoints and object keypoints to obtain keypoint embeddings. The method also includes predicting, by a hierarchical transformer encoder using the keypoint embeddings, human actions and bounding box information of when and where the human actions occur in the one or more video frames.

Abstract: A process of presenting statements captured at an incident scene. First real-time multimedia streams captured corresponding to an incident scene are converted into digital statements. Each statement is tagged to a person identity that is determined based on a biometric attribute extracted from the multimedia streams. When a responder arrives at the incident scene, second real-time multimedia streams are captured corresponding to the incident scene and an identity of a person present at the incident scene is determined. If the identity of the person determined from the second multimedia streams matches with the person identity determined from the first multimedia streams, statements tagged to the person identity are selected and a post-arrival notification is generated for the responder indicating a presence of the person at the incident scene and including the selected statements. The notification is then provided via a visual and/or audio output device operated by the responder.