Farhad Kamangar

Indoor location tracking using RSSI readings from a single Wi-Fi access point

This paper describes research towards a system for locating wireless nodes in a home environment requiring merely a single access point. The only sensor reading used for the location estimation is the received signal strength indication (RSSI) as given by an RF interface, e.g., Wi-Fi. Wireless signal strength maps for the positioning filter are obtained by a two-step parametric and measurement driven ray-tracing approach to account for absorption and reflection characteristics of various obstacles. Location estimates are then computed using Bayesian filtering on sample sets derived by Monte Carlo sampling. We outline the research leading to the system and provide location performance metrics using trace-driven simulations and real-life experiments. Our results and real-life walk-troughs indicate that RSSI readings from a single access point in an indoor environment are sufficient to derive good location estimates of users with sub-room precision.

Heterogeneous image feature integration via multi-modal spectral clustering

In recent years, more and more visual descriptors have been proposed to describe objects and scenes appearing in images. Different features describe different aspects of the visual characteristics. How to combine these heterogeneous features has become an increasing critical problem. In this paper, we propose a novel approach to unsupervised integrate such heterogeneous features by performing multi-modal spectral clustering on unlabeled images and unsegmented images. Considering each type of feature as one modal, our new multi-modal spectral clustering (MMSC) algorithm is to learn a commonly shared graph Laplacian matrix by unifying different modals (image features). A non-negative relaxation is also added in our method to improve the robustness and efficiency of image clustering. We applied our MMSC method to integrate five types of popularly used image features, including SIFT, HOG, GIST, LBP, CENTRIST and evaluated the performance by two benchmark data sets: Caltech-101 and MSRC-v1. Compared with existing unsupervised scene and object categorization methods, our approach always achieves superior performances measured by three standard clustering evaluation metrices.

A system for large vocabulary sign search

A method is presented to help users look up the meaning of an unknown sign from American Sign Language (ASL). The user submits a video of the unknown sign as a query, and the system retrieves the most similar signs from a database of sign videos. The user then reviews the retrieved videos to identify the video displaying the sign of interest. Hands are detected in a semi-automatic way: the system performs some hand detection and tracking, and the user has the option to verify and correct the detected hand locations. Features are extracted based on hand motion and hand appearance. Similarity between signs is measured by combining dynamic time warping (DTW) scores, which are based on hand motion, with a simple similarity measure based on hand appearance. In user-independent experiments, with a system vocabulary of 1,113 signs, the correct sign was included in the top 10 matches for 78% of the test queries.

Laplacian and orthogonal wavelet pyramid decompositions in coarse-to-fine registration

The authors develop orthogonal wavelet pyramid methods for registration and matching. The registration algorithm described compensates for the representations lack of translation invariance. It shows better performance when tested with Laplacian pyramids. For matching applications, they derive filters that directly generate coarse representations of candidates for comparison to library prototypes. >

Computer image analysis of ultrasound biomicroscopy of primate accommodation

PurposeTo assess and correct images of the eye for movements that can confound the evaluation of the presence, direction, and magnitude of intraocular movement of the crystalline lens equator during centrally induced ciliary muscle contraction (accommodation).MethodUltrasound biomicroscopic (UBM) video images of a cynomologus monkey crystalline lens were obtained from an independent source. The images, prior to, during, and following electrical stimulation of the Edinger-Westphal (EW) nucleus were compared for evidence of movement of the crystalline lens equator. Extraocular eye movements were assessed by use of objective computer imaging analysis techniques.ResultsExtraocular eye movements were identified and reduced by using objective computer imaging analysis techniques to register and realign the corneal images. Highly significant corrections are required to effect corneal realignment. Analysis of paired and registered images from this data source indicates that any movements of the primate lens equator are not detectable when maximum accommodation was induced by EW stimulation.ConclusionThe displacement of the edge of the primate crystalline lens equator during electrically induced contraction of the ciliary muscle is a small displacement phenomenon, only analysable after confounding extraocular movements are removed from the compared images.

A neural network for pursuit tracking inspired by the fly visual system

Abstract This paper presents an artificial neural network that detects and tracks an object moving within its field of view. This novel network is inspired by processing functions observed in the fly visual system. The network detects changes in input light intensities, determines motion on both the local and the wide-field levels, and outputs displacement information necessary to control pursuit tracking. Software simulations demonstrate the current prototype successfully follows a moving target within specified radiance and motion constraints. The paper reviews these limiting constraints and suggests future network augmentations to remove them. Despite its current limitations, the existing prototype serves as a solid foundation for a future network that promises to provide machines with the improved abilities to do high-speed pursuit tracking, interception, and collision avoidance.

Geometric feature-based matching in stereo images

Presents a geometric feature-based matching approach to solve the stereo correspondence problem. The first distinct features are extracted in a pair of stereo images using a feature extractor. Then a newly developed window-based feature point detector is used to detect feature points from the extracted feature in both images. Feature points are connected with two points to form a straight line in both images. A match function representing the requirements of the epipolar and disparity constraints in both images is proposed for straight line matching. After that straight-line correspondence is established using the match function values in the left image and corresponding ones in the right image. Feature points are ordered in both images using a geometric transformation and stored in a database in increasing order. Triplets of ordered matched points are used to construct a model polygon in the left image. Then the entire right image is searched by an exhaustive search method to find a matching polygon. A disparity test is applied to ensure that no matched feature point in an image can be rematched with another feature point. Experimental results indicate that the method performs well for real stereo images and it is suitable for many applications.

The role of dictionary learning on sparse representation-based classification

This paper analyzes the role of dictionary selection in Sparse Representation-based Classification (SRC). While SRC introduces interesting results in the field of classification, its performance is highly limited by the number of training samples to form the classification matrix. Different studies addressed this issue by using a more compact representation of the training data in order to achieve higher classification speed and accuracy. Representative selection methods which are analyzed in this paper include Metaface dictionary learning, Fisher Discriminative Dictionary Learning (FDDL), Sparse Modeling Representative Selection (SMRS), and random selection of the training samples. The first two methods build their own dictionaries via an optimization process while the other two methods select the representatives directly from the original training samples. These methods, along with the original method which uses all training samples to form the classification matrix, were examined on two face datasets and one digit dataset. The role of feature extraction was also studied using two dimensionality reduction methods, down-sampling and random projection. The results show that the FDDL method leads to the best classification accuracy followed by the SMRS method as the second best. On the other hand, the SMRS method requires a much smaller learning time which makes it more appropriate for dynamic situations where the dictionary is regularly updated with new samples. The accuracy of the Metaface dictionary learning method was specifically less than the other two methods. As expected, using all the training samples as the dictionary resulted in the best recognition rates in all the datasets but the classification times for this approach were far larger than the required time using any of the three dictionary learning methods.

Planar Curve Representation and Matching

In this paper, we discuss a method for representing and matching planar curves. The technique is based on using calculations from concentric circles to represent each curve by two sets of angles. The angles are defined by vectors constructed from the center point of the circles and the points on the curve trace that intersect each circle. The circles have incrementally increasing radii represented by the minimum radius and the radius increment value. The number of circles used specifies the level of abstraction at which the curves are represented. This representation is invariant to translation and rotation transformations. Experiments with different classes of curves have shown that our technique is robust to digitization errors and noise effects, and can perform well when the number of concentric circles are relatively small. In particular, we describe the potential applicability of this technique to fingerprint identification problem.

Reliability analysis of CSP specifications - A new method using Petri nets

Theoretical models like CSP and CCS describe computation using synchronization. Such models define independent system entities or processes that cooperate by explicit communication. In safety critical systems these communications represent visible actions which, if they do not occur or are delayed beyond their deadline, will cause a failure to occur. This paper describes the basic methodology for converting a formal description of a system into the information needed to predict system behavior as a function of observable parameters. Currently under development is a tool to permit stochastic analyses of CSP-based system specifications. The CSP-based grammar used by this tool is presented and isomorphisms between CSP-based specifications and Petri net-based stochastic models are shown. A brief example of the translation between these two formalisms is given along with (1) an analytical derivation of timing failure probability and cost minimization, and (2) discrete and continuous time Markovian analysis which provide reliability predictions for candidate designs. The translation process is currently being automated.