Verizon Communications Inc.
Systems and methods for training and validating a computer vision model for geospatial imagery

Abstract:

An exemplary geospatial image processing system generates, based on multiple detections of an object of interest detected by a computer vision model in multiple, correlated images of a geospatial location captured from different camera viewpoints, user interface content that includes a visual indication of the detected object of interest superimposed at an object position on a view of the geospatial location. The system provides the user interface content for display in a graphical user interface view of a user interface and provides, by way of the user interface, a user interface tool configured to facilitate user validation of one or more of the multiple detections of the object of interest. The system may receive, a user validation of one or more of the multiple detections of the object of interest and may train the computer vision model based on the user validation. Corresponding methods and systems are also disclosed.