Abstract: A method for natural language-based interaction with a vision-based monitoring system. The method includes obtaining a request input from a user, by the vision-based monitoring system. The request input is directed to an object detected by a classifier of the vision-based monitoring system. The method further includes obtaining an identifier associated with the request input, identifying a site of the vision-based monitoring system from a plurality of sites, based on the identifier, generating a database query, based on the request input and the identified site, and obtaining, from a monitoring system database, video frames that relate to the database query. The video frames include the detected object. The method also includes providing the video frames to the user.