Given a NITF-format satellite image and a 3D model of an object, MAGIC ATR will return high confidence object detections with latitude/longitude coordinates and headings. This wide area search analytic tool can be used: on poor quality imagery where machine learning techniques fails, to detect targets when sparse data sets exist for training of machine learning algorithms, in conjunction with machine learning algorithms to reduce false positive detections or deception of machine learning techniques
MAGIC ATR can be used out-of-the-box on panchromatic (EO) satellite imagery and the user only needs to supply a correctly dimensioned 3D model of the object of interest. MAGIC ATR allows the user to limit the tool responses to high, medium and low-confidence calls and then only has to view the candidate detections in order of confidence rank. For every candidate detection, MAGIC ATR provides latitude/longitude/elevation location, object pose, confidence and a predicted model overlay on the image chip. Analysts can view and validate the detections within minutes.
The user interacts with MAGIC ATR via a Command Line Interface (CLI) and can easily create workflows using script languages such as Python. MAGIC ATR generates KML standard outputs that can be imported to GoogleEarth, SOCET GXP, and other electronic light tables. MAGIC ATR has been designed to help analysts perform object detection in EO satellite imagery immediately without training imagery.