Lockheed Martin Space, one of the four business areas within Lockheed Martin, is headquartered in Littleton, Colorado. The organization designs, builds, and operates spacecraft and satellite systems that support space discovery and exploration.

Space domain awareness, the ability to detect and analyze objects in space, is a key problem for Lockheed Martin Space. Prior to a satellite launch, extensive testing ensures that the system works properly under a variety of conditions. Much of this testing evaluates computer vision algorithms using synthetic imagery.

Engineers use synthetic imagery and sensor data to validate that these algorithms perform as expected across diverse conditions. However, the infrastructure to set up and test these algorithms requires a high upfront cost.

Our Networked CV Algorithm Testing Harness reduces these high testing costs by providing a simple and unified environment for testing computer vision algorithms on synthetic space imagery.

Our application enables users to execute multiple algorithm types on image sequences that a centralized server provides. The client provides real-time playback with an overlaid estimate of the space object’s position. The server computes results of position, range, and orientation to compare with the real data to evaluate algorithm performance.

Our software also enables streaming of the algorithm results in real time to other users, enabling Lockheed Martin Space engineers to share critical results quickly and easily.

The system back end is a FastAPI REST with a PostgreSQL database for datasets and algorithm results. UDP streaming transmits data between the server and client, while control and event messaging occurs over HTTP and WebSocket. Algorithm processing uses OpenCV, NumPy and PySLAM. The user interface uses PyQT for real-time data visualization.