SITIS Archives  Topic Details 
Program:  SBIR 
Topic Num:  AF071233 (AirForce) 
Title:  Efficiently Computing and/or Compensating for Object Variability for Automatic Target Recognition (ATR) Applications 
Research & Technical Areas:  Sensors, Electronics 
Objective:  Mitigate the cost of simulating object variability for automatic target recognition ATR algorithms

Description:  Objects, especially ground objects, tend to have a lot of variability (objects added or removed, articulation, etc.). The traditional method for characterizing how variability affects scattering, and thereby the automatic target recognition ATR algorithm, is by developing a number of different instances of the model, computing the results for each instance, and then processing this data. This traditional method tends to be resource intensive to the point of being cost prohibitive for a study of any relevant size. An innovative approach needs to be developed that can allow a less resource intensive method for computing the scattering from variable objects for radarbased ATR.
An alternative approach is to develop ATR algorithms that have the ability to compensate for object variability without losing the ability to differentiate between object types (e.g., differentiate between an Escalade and a Hummer, regardless of whether either has anything removable attached to the outer body). This approach may just utilize various ATR algorithm techniques or it may include inverse scattering techniques also.

PHASE I: Develop an innovative method for computing the scattering from objects with a high degree of variability and an ATR algorithm that can compensate for object variability or one of the two. Determine the parameters over which the method or methods are applicable.
 
PHASE II: Further refine the method or methods developed in Phase I. Show that the method works for a dataset of reasonable size.
 
DUAL USE COMMERCIALIZATION: Military application: All ATR systems have to deal with variability, so there will be many militarybased ATR applications for this work. Commercial application: Rather than just detect obstructions, automobiles could be fitted with the capability to detect and identify deer or other moving obstructions that the driver should avoid.
 
References:  1. Zelnio, E., Garber, F., Westerkamp, L., Worrell, S., Westerkamp, J., Jarratt, M., Deardorf, C., and Ryan, P., “Characterization of ATR Systems,” Proceedings of SPIE, Vol. 3070, July 1997, pp. 223234.
2. Liu, Z., Jiangqi, H., Xie, Y., Sullivan, A., and Carin, L., “Multilevel Fast Multipole Algorithm for General Targets on a HalfSpace Interface,” IEEE Transactions on Antennas and Propagation, Vol. 50, No. 12, December 2002, pp. 18381849.

Keywords:  computational electromagnetics (CEM), automatic target recognition (ATR), CID technology, variability, inverse scattering 
Questions and Answers: 
Q: If I understand the problem correctly, you are interested in how the incident EM field will propagate after being scattered by the ground objects. 
A: 1. Overall, this topic was designed to address the issue of decreasing the sensitivity of an ATR algorithm to the variations of different instantiations of the same object (e.g., a pristine object versus an object with dents or partially/fully open doors) while retaining the ability to differentiate between different objects (e.g., the example in the solicitation, Escalade versus a Hummer). The EM portion of this solicitation is the efficient computation of the different instantiations of the same object. I'm not sure if this answers the question, but I'm not entirely sure how to interpret the question. How you go about solving the details of the above is up to you. 
Q: 1. In the EM part, is the bulk of the computational complexity in calculating the propagation of the scattered field from each instantiation? If so, would you be interested in an optimized EMonly (no ATR) approach that would require just a fraction of the entire computation to be repeated as different objects and instantiations are swapped but the propagation path/terrain remains constant? 
A: 1. At least for the phase one, the main interest for the EM portion of the problem is to accurately and efficiently compute the scattering from the different instantiations of the object of interest rather than the scattering from objects in different environments. If efficiently and accurately computing the instantiations is successful then adding the environment would be one of the next steps. When the environment is added, it should be properly coupled to the object of interest. 
Q: If I understand the problem correctly, you are interested in how the incident EM field will propagate after being scattered by the ground objects. 
A: 1. Overall, this topic was designed to address the issue of decreasing the sensitivity of an ATR algorithm to the variations of different instantiations of the same object (e.g., a pristine object versus an object with dents or partially/fully open doors) while retaining the ability to differentiate between different objects (e.g., the example in the solicitation, Escalade versus a Hummer). The EM portion of this solicitation is the efficient computation of the different instantiations of the same object. I'm not sure if this answers the question, but I'm not entirely sure how to interpret the question. How you go about solving the details of the above is up to you. 
Q: 1. In the EM part, is the bulk of the computational complexity in calculating the propagation of the scattered field from each instantiation? If so, would you be interested in an optimized EMonly (no ATR) approach that would require just a fraction of the entire computation to be repeated as different objects and instantiations are swapped but the propagation path/terrain remains constant? 
A: 1. At least for the phase one, the main interest for the EM portion of the problem is to accurately and efficiently compute the scattering from the different instantiations of the object of interest rather than the scattering from objects in different environments. If efficiently and accurately computing the instantiations is successful then adding the environment would be one of the next steps. When the environment is added, it should be properly coupled to the object of interest. 
