US 11,941,902 B2
System and method for asset serialization through image detection and recognition of unconventional identifiers
Charles Scott McAllister, Silver Spring, MD (US); Arun Aggarwal, Dallas, TX (US); John Henry Rudisill, Atlanta, GA (US); Viral Chawda, Frisco, TX (US); and Kimball Hill, Salt Lake City, UT (US)
Assigned to KPMG LLP, New York, NY (US)
Filed by KPMG LLP, New York, NY (US)
Filed on Apr. 14, 2022, as Appl. No. 17/659,286.
Claims priority of provisional application 63/265,167, filed on Dec. 9, 2021.
Prior Publication US 2023/0186662 A1, Jun. 15, 2023
Int. Cl. G06V 30/14 (2022.01); G06V 30/12 (2022.01); G06V 30/19 (2022.01); G06V 30/224 (2022.01)
CPC G06V 30/1456 (2022.01) [G06V 30/133 (2022.01); G06V 30/19147 (2022.01); G06V 30/1916 (2022.01); G06V 30/2247 (2022.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented system for unique asset serialization, the system comprising:
an interactive user interface that is configured to receive one or more inputs;
a database interface that communicates with a database that stores and manages asset data; and
a processor executing on a mobile device and coupled to the interface and the database interface, the processor further configured to perform the steps of:
receiving, via an input interface, one or more images of a unique asset;
detecting, via a computer vision detection model, one or more unique serial number candidates on the unique asset in the received one or more images of the unique asset;
performing, via a prediction model, text recognition on the one or more unique serial number candidates and identifying one or more predicted unique serial numbers with corresponding confidence levels;
performing post processing on the one or more predicted unique serial numbers to improve prediction accuracy;
displaying, via the interactive user interface executing on the mobile device, the one or more predicted unique serial numbers;
receiving one or more user inputs responsive to the one or more predicted unique serial numbers; and
improving the prediction model based on the one or more user inputs.