This application involves a close collaboration with the Acoustics and Signal Processing Branch of the Army Research Laboratory (ARL). It revolves around the use of a software architecture for an expandable testbed sensor tracking system utilizing a remotely netted acoustic sensor array system that detects, tracks, and identifies ground and air targets at tactical ranges. In the process of building this system, a testbed has been designed to evaluate newly developed algorithms in real time. The test bed architecture is modular and allows the output of various sensor systems (acoustic, seismic, optical, GPS) to interface with a gateway that correlates and processes the information, and in turn displays results and statistics in real time. The test-bed is fully operational with dual RF communication between the operator and the sensors in the field.
A detailed description of this testbed can be found at [.Choy 1995.], together with a discussion of the overall system concept, and a detailed breakdown of each software function. A description of processing algorithms and data structures used is also available along with recommendations for specific implementation methodologies. The general design philosophy of the testbed architecture is to target basic operational functions and isolate each function as an independent software task with well defined inputs and outputs. Each package will manage its own private memory context and should (in order to help promote software independence) ultimately consist of re-entrant code. The expandable testbed is a tool for research and development, hence it has built-in mechanisms for adjusting and tuning the system during its operational state. This allows for on-the-fly experimentation with new software algorithms and hardware devices without having to re-build the system each time from the ground up. It also enables diagnostics to be performed on the system as it is being developed so that the developer can make intelligent decisions about the functional validity of the software and hardware being tested. Data generated by the overall system are sufficiently complete to allow for post-analysis of the tracking algorithm performance levels. The information generated for post-processing can be different from the information generated for real time display.
We plan to develop and test auditory-based algorithms for the identification and recognition of vehicles and aircrafts based on their acoustic signals. These algorithms would eventually be incorporated as part of the gateway data fusion processes where various tracking algorithms are used to correlate and process the incoming information from multiple sensor arrays. The preliminary auditory algorithms developed and tested so far have been primarily composed of self-normalized spectral estimates of the acoustic targets followed by a cortical multiscale representation of the spectra. Details of the speed and nature of different vehicles could be readily distinguished across different scales of the representation.
The Acoustics and Signal Processing Branch at ARL is currently investing considerable effort developing algorithms that can consistently provide greater than 90% probability of correct target identification. Of issue is the efficient separation of target signature from the background clutter. Our auditory algorithms have the potential to increase the capability to identify targets with high confidence and low false alarm. This effort will be significantly aided by ARL's unique database resources consisting of extensive recordings of target vehicles in realistic environments which are essential to develop a fully automated system.