Surveillance system with human behavior prediction by human action recognition
AssigneeSensormatic Electronics Corporation
InventorsSantle Camilus, Manjuprakash Rama Rao
Life of this patent
Estimating the term…
Abstract
The present invention concerns surveillance systems that flag the potential threats automatically using intelligent systems. It can then notify or automatically alert the security personnel of impending dangers. Such a system can lower the cognitive load on the security personnel and can assist them to bring to prioritize their attention to potential threats and thereby improve the overall efficiency of the system. There could also be savings in labor cost.
Claims20
A method for analyzing surveillance data from a surveillance system, the method comprising:
- a neural network recognizing actions performed by one or more individuals based on analyzing surveillance feeds from the surveillance system;
- a support vector machine (SVM) classifier categorizing the recognized actions as having a type of safe, less safe, or unsafe and generating vulnerability scores indicating whether the recognized actions are determined to be harmful or unusual based on the types of the recognized actions, duration of the recognized actions, and sequences of recognized actions of different types;
- and
- generating alerts in response to determining that the vulnerability score is above a predetermined threshold.
A system for analyzing surveillance data from a surveillance system, the system comprising:
- a neural network for recognizing actions performed by one or more individuals based on analyzing surveillance feeds from the surveillance system;
- and
- a support vector machine (SVM) classifier for categorizing the recognized actions as having a type of safe, less safe, or unsafe and generating vulnerability scores indicating whether the recognized actions are determined to be harmful or unusual based on the types of the recognized actions, duration of the recognized actions, and sequences of recognized actions of different types and for generating alerts in response to determining that the vulnerability score is above a predetermined threshold.
Dependencies are inferred from the claim text.