Bayesian Contrast Measures and Clutter Distribution Determinants of Human Target Detection.

Human target detection is known to be dependent on a number of components: one, basic electro-optics including image contrast, the target size, pixel resolution, and contrast sensitivity; two, target shape, image type and features, types of clutter; and three, context and task requirements. Here, we consider a Bayesian approach to investigating how these components contribute to target detection. To this end, we develop and compare three different formulations for contrast: mean contrast, perceptual contrast, and a Bayesian-based histogram contrast statistic. Results on past detection data show how the latter contrast measure correlates well with human performance factoring out all other dimensions. As for clutter, our findings show that with large targets, there are effectively no clutter effects. Furthermore, clutter does not have a major effect on detection when it is not contiguous with the target even when it is smaller. However, except for large targets, when the target is contiguous with the clutter, detection clearly decreases as a function of the similarity of target and clutter features-creating type of "clutter camouflage". This Bayesian formulation uses priors based on the contrast histogram statistics derived from all the images, the image context, and implies that human observers have adapted their criteria to fit with the image set, context, and task.

A syntactic two-component encoding model for the trajectories of human actions.

Human actions have been widely studied for their potential application in various areas such as sports, pervasive patient monitoring, and rehabilitation. However, challenges still persist pertaining to determining the most useful ways to describe human actions at the sensor, then limb and complete action levels of representation and deriving important relations between these levels each involving their own atomic components. In this paper, we report on a motion encoder developed for the sensor level based on the need to distinguish between the shape of the sensor's trajectory and its temporal characteristics during execution. This distinction is critical as it provides a different encoding scheme than the usual velocity and acceleration measures which confound these two attributes of any motion. At the same time, we eliminate noise from sensors by comparing temporal and spatial indexing schemes and a number of optimal filtering models for robust encoding. Results demonstrate the benefits of spatial indexing and separating the shape and dynamics of a motion, as well as its ability to decompose complex motions into several atomic ones. Finally, we discuss how this specific type of sensor encoder bears on the derivation of limb and complete action descriptions.

Monitoring and Analysis of Respiratory Patterns Using Microwave Doppler Radar.

Noncontact detection characteristic of Doppler radar provides an unobtrusive means of respiration detection and monitoring. This avoids additional preparations, such as physical sensor attachment or special clothing, which can be useful for certain healthcare applications. Furthermore, robustness of Doppler radar against environmental factors, such as light, ambient temperature, interference from other signals occupying the same bandwidth, fading effects, reduce environmental constraints and strengthens the possibility of employing Doppler radar in long-term respiration detection, and monitoring applications such as sleep studies. This paper presents an evaluation in the of use of microwave Doppler radar for capturing different dynamics of breathing patterns in addition to the respiration rate. Although finding the respiration rate is essential, identifying abnormal breathing patterns in real-time could be used to gain further insights into respiratory disorders and refine diagnostic procedures. Several known breathing disorders were professionally role played and captured in a real-time laboratory environment using a noncontact Doppler radar to evaluate the feasibility of this noncontact form of measurement in capturing breathing patterns under different conditions associated with certain breathing disorders. In addition to that, inhalation and exhalation flow patterns under different breathing scenarios were investigated to further support the feasibility of Doppler radar to accurately estimate the tidal volume. The results obtained for both experiments were compared with the gold standard measurement schemes, such as respiration belt and spirometry readings, yielding significant correlations with the Doppler radar-based information. In summary, Doppler radar is highlighted as an alternative approach not only for determining respiration rates, but also for identifying breathing patterns and tidal volumes as a preferred nonwearable alternative to the conventional contact sensing methods.

Non-contact measurement of respiratory function and deduction of tidal volume.

This paper further the investigation of Doppler radar feasibility in measuring the flow in and out due to inhalation and exhalation under different conditions of breathing activities. Three different experiment conditions were designed to investigate the feasibility and consistency of Doppler radar which includes the combination of the states of normal breathing, deep breathing and apnoea state were demonstrated. The obtained Doppler radar signals were correlated and compared with the gold standard medical device, spirometer, yielding a good correlations between both devices. We also demonstrated the calibration of the Doppler radar signal can be performed in a simple manner in order to have a good agreements with the spirometer readings. The measurement of the flow in and out during the breathing activities can be measured accurately under different dynamics of breathing as long as the calibration is performed correctly.

Multi-kinect skeleton fusion for physical rehabilitation monitoring.

Kinect has been increasingly applied in rehabilitation as a motion capture device. However, the inherent limitations significantly hinder its further development in this important area. Although a number of Kinect fusion approaches have been proposed, only a few of them was actually considered for rehabilitation. In this paper, we propose to fuse information from multiple Kinects to achieve this. Given the specific scenario of users suffering from limited range of movements, we propose to calibrate depth cameras in multiple Kinects with 3D positions of joints on a human body rather than in a checkerboard pattern, so that patients are able to calibrate Kinects without extra support. Kalman filter is applied for skeleton-wise Kinect fusion since skeleton data (3D positions of joints) and its derivatives are preferred by physiotherapists to evaluate the exercise performance of patients. Various preliminary experiments were conducted to illustrate the accuracy of proposed calibration and fusion approach by comparing with a commercial Vicon system®, confirming the practical use of the system in rehabilitation exercise monitoring.

Functional range of movement of the hand: declination angles to reachable space.

The measurement of the range of hand joint movement is an essential part of clinical practice and rehabilitation. Current methods use three finger joint declination angles of the metacarpophalangeal, proximal interphalangeal and distal interphalangeal joints. In this paper we propose an alternate form of measurement for the finger movement. Using the notion of reachable space instead of declination angles has significant advantages. Firstly, it provides a visual and quantifiable method that therapists, insurance companies and patients can easily use to understand the functional capabilities of the hand. Secondly, it eliminates the redundant declination angle constraints. Finally, reachable space, defined by a set of reachable fingertip positions, can be measured and constructed by using a modern camera such as Creative Senz3D or built-in hand gesture sensors such as the Leap Motion Controller. Use of cameras or optical-type sensors for this purpose have considerable benefits such as eliminating and minimal involvement of therapist errors, non-contact measurement in addition to valuable time saving for the clinician. A comparison between using declination angles and reachable space were made based on Hume's experiment on functional range of movement to prove the efficiency of this new approach.

Further applications of Doppler radar for non-contact respiratory assessment.

This paper further investigates the use of Doppler radar for detecting and identifying certain human respiratory characteristics from observed frequency and phase modulations. Specifically, we show how breathing frequencies can be determined from the demodulated signal leading to identifying abnormalities of breathing patterns using signal derivatives, optimal filtering and standard statistical measures. Specifically, we report results on a robust method for distinguishing cessation of the normal breathing cycle. The proposed approach can have potential application in the management of sudden infant death syndrome(SIDS) and sleep apnea.

Real-time discriminative background subtraction.

The authors examine the problem of segmenting foreground objects in live video when background scene textures change over time. In particular, we formulate background subtraction as minimizing a penalized instantaneous risk functional--yielding a local online discriminative algorithm that can quickly adapt to temporal changes. We analyze the algorithm's convergence, discuss its robustness to nonstationarity, and provide an efficient nonlinear extension via sparse kernels. To accommodate interactions among neighboring pixels, a global algorithm is then derived that explicitly distinguishes objects versus background using maximum a posteriori inference in a Markov random field (implemented via graph-cuts). By exploiting the parallel nature of the proposed algorithms, we develop an implementation that can run efficiently on the highly parallel graphics processing unit (GPU). Empirical studies on a wide variety of datasets demonstrate that the proposed approach achieves quality that is comparable to state-of-the-art offline methods, while still being suitable for real-time video analysis ( ≥ 75 fps on a mid-range GPU).

The application of remote sensing for detecting mass graves: an experimental animal case study from Costa Rica.

Detection of mass graves utilizing the hyperspectral information in airborne or satellite imagery is an untested application of remote sensing technology. We examined the in situ spectral reflectance of an experimental animal mass grave in a tropical moist forest environment and compared it to an identically constructed false grave which was refilled with soil, but contained no cattle carcasses over the course of a 16-month period. The separability of the in situ reflectance spectra was examined with a combination of feature selection and five different nonparametric pattern classifiers. We also scaled up the analysis to examine the spectral signature of the same experimental mass grave from an air-borne hyperspectral image collected 1 month following burial. Our results indicate that at both scales (in situ and airborne), the experimental grave had a spectral signature that was distinct and therefore detectable from the false grave. In addition, we observed that vegetation regeneration was severely inhibited over the mass grave containing cattle carcasses for up to a period of 16 months. This experimental study has demonstrated the real utility of airborne hyperspectral imagery for the detection of a relatively small mass grave (5 m(2)) within a specific climatic zone. Other climatic zones will require similar actualistic modeling studies, but it is clear that the applications of this technology provide the international community with both an early detection tool and a tool for ongoing monitoring.

Structural and view-specific representations for the categorization of three-dimensional objects.

It has been debated whether object recognition depends on structural or view-specific representations. This issue is revisited here using a paradigm of priming, supervised category learning, and generalization to novel viewpoints. Results show that structural representations can be learned for three-dimensional (3D) objects lacking generalized-cone components (geons). Metric relations between object parts are distinctive features under such conditions. Representations preserving 3D structure are learned provided prior knowledge of object shape and sufficient image input information is available; otherwise view-specific representations are generated. These findings indicate that structural and view-specific representations are related through shifts of representation induced by learning.

Graphical models and point pattern matching.

This paper describes a novel solution to the rigid point pattern matching problem in Euclidean spaces of any dimension. Although we assume rigid motion, jitter is allowed. We present a noniterative, polynomial time algorithm that is guaranteed to find an optimal solution for the noiseless case. First, we model point pattern matching as a weighted graph matching problem, where weights correspond to Euclidean distances between nodes. We then formulate graph matching as a problem of finding a maximum probability configuration in a graphical model. By using graph rigidity arguments, we prove that a sparse graphical model yields equivalent results to the fully connected model in the noiseless case. This allows us to obtain an algorithm that runs in polynomial time and is provably optimal for exact matching between noiseless point sets. For inexact matching, we can still apply the same algorithm to find approximately optimal solutions. Experimental results obtained by our approach show improvements in accuracy over current methods, particularly when matching patterns of different sizes.

Component optimization for image understanding: a Bayesian approach.

In this paper, the optimizations of three fundamental components of image understanding: segmentation/annotation, 3D sensing (stereo) and 3D fitting, are posed and integrated within a Bayesian framework. This approach benefits from recent advances in statistical learning which have resulted in greatly improved flexibility and robustness. The first two components produce annotation (region labeling) and depth maps for the input images, while the third module integrates and resolves the inconsistencies between region labels and depth maps to fit most likely 3D models. To illustrate the application of these ideas, we have focused on the difficult problem of fitting individual tree models to tree stands which is a major challenge for vision-based forestry inventory systems.

An eigenspace projection clustering method for inexact graph matching.

In this paper, we show how inexact graph matching (that is, the correspondence between sets of vertices of pairs of graphs) can be solved using the renormalization of projections of the vertices (as defined in this case by their connectivities) into the joint eigenspace of a pair of graphs and a form of relational clustering. An important feature of this eigenspace renormalization projection clustering (EPC) method is its ability to match graphs with different number of vertices. Shock graph-based shape matching is used to illustrate the model and a more objective method for evaluating the approach using random graphs is explored with encouraging results.

Development of configural 3D object recognition.

There is evidence for the late development in humans of configural face and animal recognition. We show that the recognition of artificial three-dimensional (3D) objects from part configurations develops similarly late. We also demonstrate that the cross-modal integration of object information reinforces the development of configural recognition more than the intra-modal integration does. Multimodal object representations in the brain may therefore play a role in configural object recognition.

Towards a decision support system for health promotion in nursing.

AIMS: This study was designed to investigate what type of models, techniques and data are necessary to support the development of a decision support system for health promotion practice in nursing. Specifically, the research explored how interview data can be interpreted in terms of Concept Networks and Bayesian Networks, both of which provide formal methods for describing the dependencies between factors or variables in the context of decision-making in health promotion. BACKGROUND: In nursing, the lack of generally accepted examples or guidelines by which to implement or evaluate health promotion practice is a challenge. Major gaps have been identified between health promotion rhetoric and practice and there is a need for health promotion to be presented in ways that make its attitudes and practices more easily understood. New tools, paradigms and techniques to encourage the practice of health promotion would appear to be beneficial. Concept Networks and Bayesian Networks are techniques that may assist the research team to understand and explicate health promotion more specifically and formally than has been the case, so that it may more readily be integrated into nursing practice. METHODS: As the ultimate goal of the study was to investigate ways to use the techniques described above, it was necessary to first generate data as text. Textual descriptions of health promotion in nursing were derived from in-depth qualitative interviews with nurses nominated by their peers as expert health promoting practitioners. FINDINGS: The nurses in this study gave only general and somewhat vague outlines of the concepts and ideas that guided their practice. These data were compared with descriptions from various sources that describe health promotion practices in nursing, then examples of a Conceptual Network and a representative Bayesian Network were derived from the data. CONCLUSIONS: The study highlighted the difficulty in describing health promotion practice, even among nurses recognized for their expertise in health promotion. Nevertheless, it indicated the data collection and analysis methods necessary to explicate the cognitive processes of health promotion and highlighted the benefits of using formal conceptualization techniques to improve health promotion practice.

Neural computations of algebraic and geometrical structures.

How Artificial Neural Networks (ANN) can be used to solve problems in algebra and geometry by modelling specific subnetwork nodes and connections is considered. This approach has the benefit of producing ANNs with well-defined hidden units and reduces the search to parameters which satisfy known model constraints-yet still gains from the benefits inherent in neural computing architectures.