Aided and automatic target recognition based upon sensory inputs from image forming systems

TLDR

Automatic target recognition (ATR) is used in the military to discriminate targets, with most work on human‑aided systems but growing interest in autonomous methods and multisensor fusion/model‑based algorithms. The paper reviews a decade of Army Laboratory research on object‑recognition algorithms, processors, and evaluation methods, detailing definitions, performance metrics, and state‑of‑the‑art results for multisensor fusion and model‑based ATR techniques. Performance data indicate that ATR can achieve useful accuracy, with signal‑to‑noise and clutter affecting results, and the study highlights the need for image‑science advances to further improve real‑world performance.

Abstract

This paper systematically reviews 10 years of research that several Army Laboratories conducted in object recognition algorithms, processors, and evaluation techniques. In the military, object recognition is applied to the discrimination of military targets, ranging from human-aided to autonomous operations, and is called automatic target recognition (ATR). The research described here has been concentrated in human-aided target recognition applications, but some attention has been paid to automatic processes. Definitions and performance metrics that have been developed are described along with performance data showing the present state-of-the-art. The effects of signal-to-noise and clutter parameters are indicated in the data. Multisensor fusion and model-based algorithms are discussed as the latest techniques under consideration by the military research community. The results demonstrate that useful performance can be achieved, and tools are evolving to understand and improve the performance under real-world conditions. The referenced research strongly indicates the need for the development of image science, as described in the paper, to support the theoretical underpinnings of ATR.