2019_programme: ATR PERFORMANCE FOR TARGET CLUSTERING
- Session: 04. Towards Automatic Target Recognition. Detection, Classification and Modelling
Organiser(s): Groen Johannes, Jans Wolfgang, Pailhas Yan, Myers Vincent
- Lecture: ATR PERFORMANCE FOR TARGET CLUSTERING [invited]
Paper ID: 874
Author(s): Melo José, Dugelay Samantha
Presenter: Melo José
Presentation type: oral
Abstract: The goal of Target Clustering algorithms is to cluster multiple detections of the same target, in a way that can minimize the uncertainty in their location. In order to do so, clustering algorithms usually build on data provided by an ATR. In particular, they use information on the location of a given detection in a tile, together with an associated weight highlighting how likely a given ATR detection is to be originated from a target.\n\nIn this article we will present an analysis and characterization of the performance of the ATR under use by the CMRE MCM group, always with the viewpoint of an input to the target clustering algorithm. Motivations for this are twofold. First, it has been observed that in some situations the ATR can fail to completely detect the target. On the other hand, the ATR can sometimes detect a target but assign it a low score which then, depending on the acceptance threshold, can cause the target to remain undetected. Therefore, the output of the ATR needs to be characterized. The analysis will build on collected experimental data, collected during past trials which cover different operational scenarios and environmental conditions, and curated by an expert.\n\nResults from this analysis will enable to derive an empirically driven probability of detection, but also to statistically characterize the distribution of the ATR scores under different scenarios. It is expected that the obtained results will enable an accurate simulation of an unmanned MCM mission. Moreover, such results will provide a better link between ATR and clustering algorithm.
Download the full paper
- Corresponding author: Mr Melo Jose
Affiliation: NATO CMRE
Country: Italy
e-mail: