The use of advanced signal processing techniques such as signal recovery, adaptive beamforming and beam optimization solutions are explored to realize the potential for networked acoustic sensors to detect, track and classify target vehicles. The ability to determine the number and types of vehicles, when closely spaced, is a particularly challenging motivation. A reasonable cost effective solution is sought using acoustic arrays less than four (4) feet across and containing less then eight (8) microphones. Our approach addresses the development of algorithm enhancements, which are additive to the total solution. Unique to this approach is the application of Technology Engineering Research Inc (TERI) previously conducted research in the use of beamforming methods to detect and track human speakers in a crowded environment for the purpose of speech recognition and speaker identification. These novel techniques use blind source separation, deconvolution, and maximization algorithms that can determine and extract the multiple target signatures within the beam to enhance existing classification algorithms and diminish background environmental effects. Phase I will be used to review presently employed classification and beamforming techniques and how the proposed new algorithms will build and enhance these existing solutions. Improvements in blind separation of targets and advanced adaptive algorithms are being developed by TERI for use in speech recognition applications. The outcome of this effort will be to provide a combination of improved beamforming methods with blind separation technologies, leading to applications, which will enable improvements in the ability to provide detailed surveillance of individual targets in a cluttered multi-target environment. The algorithmic approach will ultimately become a module as part of a top level algorithm with the purpose of identifying the location of a tactically significant target formation, through the use of various counting, tracking, and classification sub-algorithms. TERI will submit MatLab algorithm and simulation of multiple target classification improvements to the Government. Improved blind separation algorithms will be integrated as part of the Government acoustic data acquisition and algorithm development environment, then tested in the field for enhanced multiple target identification and tracking during Phase II. Applications for this technology exist for extended range and enhanced detection accuracy and effectiveness wherever intelligent acoustic sensors are employed under variable environmental conditions. Use for tracking acoustic signatures can be applied to speaking participants during a tele-conference to detect and "focus" on comments, for mobile phones, for use in tele-medicine under noisy conditions and for unmanned aerial surveillance vehicles.
