Abstract
This disclosure describes techniques that improve the recognition of a person observed in video footage captured by security cameras by leveraging a large number of images of the person obtained from various perspectives from multiple cameras. Images of a person moving through a given space and appearing at various angles and look-directions are obtained and clustered. The cluster of images of a person is subjected to face detection to obtain a gallery of face images for the person. To recognize the person, similarity scores are computed between a reference image of the person and each of the images in the face gallery. The similarity score is aggregated, and, if the aggregate meets a threshold, the person in the gallery of images is identified as the person in the reference image. The techniques result in more accurate and reliable recognition of individuals - an improved recall rate and a reduced false alarm rate, in turn lowering the cost of human review of security footage.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
NA, "Clustering Person Appearances to Improve Person Recognition", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/9512