US Patent:
20210350117, Nov 11, 2021
Inventors:
- CHICAGO IL, US
TERESA DANKEL - LEXINGTON MA, US
PATRICK HENSON - BEDFORD MA, US
ARAL SARRAFI - LOWELL MA, US
International Classification:
G06K 9/00
G06K 9/62
G06K 9/32
G06T 7/73
Abstract:
A method of detecting anomalous poses is provided, including: detecting an object of interest in a field of view of a video camera; generating, by a neural network, a pose vector associated with a pose of the object; comparing the pose vector to one or more clusters of pose vectors in a dataset of pose vectors, each cluster of pose vectors comprising one or more pose vectors generated from poses of objects; determining a value associated with the similarity of the pose vector to at least one of the one or more pose vectors in the cluster; comparing the value to a predetermined value; and based on the comparison of the value, adding the pose vector to the dataset or generating an alert.