Computer on Watch

An artificial intelligence application that automatically identifies objects in aerial imagery could reduce the time analysts spend in manually combing through images.
A goal of the program is to develop a system that can identify spatial relationships between objects in a scene, such as counting how many planes are parked at the terminal on the left.
A goal of the program is to develop a system that can identify spatial relationships between objects in a scene, such as counting how many planes are parked at the terminal on the left.

The research team on the Computer on Watch program is exploring methods for applying machine learning to imagery analysis. We are developing a prototype system that will accelerate the process of analyzing overhead images by summarizing the spatial and temporal relationships between objects in a scene. One of the challenges in creating these types of systems is that for many applications there is a lack of relevant data that has been sufficiently labeled to train the software. We are investigating techniques for efficiently labeling data for artificial intelligence (AI) systems in order to reduce the amount of time required to develop and improve machine learning applications. Our near-term objective is to demonstrate a system that recognizes objects in overhead imagery and highlights uncertain regions for further human analysis. A longer-term goal of the Computer on Watch program is to develop a cognitive assistant that not only automates these tasks, but also provides a natural language interface to allow imagery analysts to seamlessly interact with the system.