This paper develops an algorithm for autonomous tracking of a person (target) within a crowded and temporally dynamic scene using a multispectral imaging system. The camera is stationary, the field of view is static, and the sensor pixel footprint is on the order of one inch. The operator designates the target to be tracked by selecting a single target-pixel in the first image frame, preferably close to the center of mass of the observable portion of the target in that particular frame. Following the initial designation, the algorithm provides tracking of the target in real-time autonomously with minimal latency. The tracking algorithm is based on a novel temporally adaptive spatial-spectral filter bank used to detect target presence or lack thereof in the field-of-regard of the video frame produced by the multispectral camera. The theory of the temporally adaptive spatial-spectral filter is based on an extension of our earlier work on the enhanced matched filter bank (EMFB). The concept of EMFB is founded on the theory of spatial matched filters, which is the optimal correlation filter for detection of a known image corrupted by noise.
An essential component within most approaches used to evaluate ATR algorithm performance is an image database from
which are chosen a training set of images. Several fundamental questions arise regarding the adequacy of the database to
represent the desired domain of effectiveness, the sufficiency of the training set, potentiality of enhancing the
constituents of the training set, suitability to determine signal-to-clutter performance, and realism of fairly comparing
performance of ATR algorithms to one another. These questions have been addressed through an investigation into a
unified approach for database analysis and how it can be applied to evaluating ATR performance metrics.
Fourier correlators perform space-invariant linear filtering on all input points, so they can
identify and locate patterns in parallel. Each output point is a weighted sum of
components of the Fourier transform of the input, so the discriminants used are inherently
linear. As most practical problems are not linearly discriminable, that causes a problem.
This paper describes a quite general solution involving nonlinear combining of
nonlinearly processed outputs from multiple Fourier masks. The design of the masks and
nonlinearities allows very powerful nonlinear discrimination that preserves the space-invariant
feature that makes Fourier correlators attractive. Given a set of target-class
images, henceforth referred to as the training set or trainers, the algorithm developed
herein computes an ordered set of classifier filters - Generalized Matched Filters (GMFs)
threshold values. An unlabeled image is applied to the classifier filter set, hereafter
referred to as super-generalized matched filter (SGMF). If the peak response of any of the
classifier filters (GMFs) to the unlabeled test image exceeds the threshold level the
decision is made in favor of labeling the image as target-class otherwise it is labeled non-target-
Wireless sensors networks are currently being used in different engineering fields such as civil, mechanical and
aerospace engineering for damage detection. Each network contains approximately hundreds to thousands of MEMS
sensors that communicate to its base station. These sensors are placed in different environments and locations that create
changes in their output due to obstacles or interference between them and their base station. A research study was
conducted on wireless MEMS sensor nodes to evaluate the noise level and the effect of environmental interferences as
well as their maximum distance communication. In this paper, the effect of interference environments and obstacles
such as magnetic field created by electricity and cell phone communications, concrete and metal enclosures, and
outside/inside environments were evaluated. In addition, a neural network computer simulation was developed to learn
and teach the users what it takes to classify signals such as time, amount of samples and overtraining in order to obtain
the correct output instead of an unknown. By gathering all this information it helps to save money and time in any
application wireless MEMS sensors are used and idealized models and pictures of communication paths have been
created for easier evaluation of the MEMS sensor networks.