Yan Zhai

An Object-Tracking Algorithm Based on Multiple-Model Particle Filtering With State Partitioning

As evidenced by the recent works of many researchers, the particle-filtering (PF) framework has revolutionized probabilistic visual target tracking. In this paper, we present a new particle filter tracking algorithm that incorporates the multiple-model (MM) paradigm and the technique of state partitioning with parallel filters. Traditionally, most tracking algorithms assume that a target operates according to a single dynamic model. However, the single-model assumption can cause the tracker to become unstable, particularly when the target has complex motions and when the camera has abrupt ego-motions. In the new tracking algorithm, a target was assumed to operate according to one dynamic model from a finite set of models. The switching process from one model to another was governed by a jump Markov process. Based on the improved MM particle filter framework, we offer a new design strategy that adopts the state-partitioning technique and a bank of parallel extended Kalman filters to construct a better proposal distribution to achieve further estimation accuracy. We have conducted extensive testing for the proposed tracking algorithm, and key outcomes were given in the results section. It has been demonstrated by the experiments that this approach gave significantly improved estimations, enabling the new particle filter to effectively track human subjects.

A Three-Dimensional Swept Volume Display Based on LED Arrays

Stereoscopic, or multi-view, display systems are considered as better alternatives to conventional two-dimensional (2D) displays, since such systems can provide important visual cues for the human brain to process three-dimensional (3D) objects. An auto-stereoscopic display is a device that can render 3D images for viewers without the aid of special headgear or glasses. In this paper, we present a new design of an auto-stereoscopic swept-volume display (SVD) system based on light-emitting diode (LED) arrays. This system is constituted of a display device and a graphics control sub-system. The display device is a 2D rotating panel of LEDs, relying on “persistence of vision” to generate 3D images. The graphics control sub-system is composed of a combination of PC software, field-programmable gate arrays (FPGAs), and supporting circuitry. The primary task of the graphics control sub-system is to process 3D data and control each LED. In addition, a new 3D image generation and rendering method was developed to reduce the bandwidth requirement and to facilitate 3D image display. Demonstrated in the experiments, a prototype of this system is capable of displaying 3D images and videos with full 360 ° view angles.

A New Centralized Sensor Fusion-Tracking Methodology Based on Particle Filtering for Power-Aware Systems

In this paper, we address the problem of target tracking in a collaborative acoustic sensor network. To cope with the inherent characteristics and constraints of wireless sensor networks, we present a novel target-tracking algorithm with power-aware concerns. The underlying tracking methodology is described as a multiple-sensor tracking/fusion technique based on particle filtering. As discussed in the most recent literature, particle filtering is defined as an emerging Monte Carlo state estimation technique with proven superior performance in many target-tracking applications. More specifically, in our proposed method, each activated sensor transmits the received acoustic intensity and the direction of arrival (DOA) of the target to the sensor fusion center (a dedicated computing and storage platform, such as a microserver). The fusion center uses each received DOA to generate a set of estimations based on the state partition technique, as described later in this paper. In addition, a set of sensor weights is calculated based on the acoustic intensity received by each activated sensor. Next, the weighted sum of the estimates is used to generate the proposal distribution in the particle filter for sensor fusion. This technique renders a more accurate proposal distribution and, hence, yields more precise and robust estimations of the target using fewer samples than those of the traditional bootstrap filter. In addition, since the majority of the signal processing efficiently resides on the fusion center, the computation load at the sensor nodes is limited, which is desirable for power-aware systems. Last, the performance of the new tracking algorithm in various tracking scenarios is thoroughly studied and compared with standard tracking methods. As shown in the theory and demonstrated by our experimental results, the state-partition-based centralized particle filter reliably outperforms the traditional method in all experiments.

Efficient Adaptive Nonlinear ECHO Cancellation, Using Sub-band Implementation of the Adaptive Volterra Filter

The adaptive Volterra filter has been successfully applied in nonlinear acoustic echo cancellation (AEC) systems and nonlinear line echo cancellation systems, but its applications are limited by its required computational complexity and slow convergence rate, especially for systems with long memory length. In this paper, by leveraging a multi-channel configuration of the Volterra filter and the sampling theory for nonlinear systems, we extend linear subband delay-less adaptive filter techniques to develop an efficient sub-band implementation of the adaptive Volterra filter. The developed sub-band configuration of the adaptive Volterra filter can greatly improve the convergence rate and reduce the computational complexity of nonlinear echo cancellers, which is shown by analyses and simulations

Spectral signature calculations and target tracking for remote sensing

Enhanced tornado detection and tracking can prevent loss of life and property damage. The research weather surveillance radar (WSR)-88D locally operated by the National Severe Storms Laboratory (NSSL) in Norman, OK, has the unique capability of collecting massive volumes of time-series data over many hours, which provides a rich environment for evaluating our new postprocessing algorithms. With the advent of more memory and computing power, new state-of-the-art algorithms can be explored. In this paper, an approach of identifying tornado vortices in Doppler spectra is proposed and investigated through the use of neural networks. Once the coordinate of the tornado has been established, the research question becomes the following: Can we apply target tracking algorithms to a volume of radar data to make estimations about where the tornado is going? In recent years, particle filters have attracted great attention in several research communities. These filters are used in problems where time-varying signals must be processed in real time, and the objective is to estimate various unknowns of the signals and to detect events described by the signals. The standard solutions of such problems in many applications are based on the Kalman or extended Kalman filters. In situations when the models that describe the behavior of the system are highly nonlinear and/or the noise that distorts the signals is non-Gaussian, the Kalman-filter-based algorithms provide solutions that may be far from optimal. Here, the path of the tornado follows a path that may be described by a set of nonlinear equations. To estimate the path, the particle filter will provide the better estimates. In addition to the single WSR-88D sensor designs, data fusion and tracing designs are also given for a four-node remote sensor network in central Oklahoma. By incorporating the data from each of the sensors, improvements in tracking are illustrated. The particle-filtering algorithms are especially effective in a networked system of sensors when they are in a data-fusion setting.

Adaptive Kalman Filtering for Histogram-Based Appearance Learning in Infrared Imagery

Targets of interest in video acquired from imaging infrared sensors often exhibit profound appearance variations due to a variety of factors, including complex target maneuvers, ego-motion of the sensor platform, background clutter, etc., making it difficult to maintain a reliable detection process and track lock over extended time periods. Two key issues in overcoming this problem are how to represent the target and how to learn its appearance online. In this paper, we adopt a recent appearance model that estimates the pixel intensity histograms as well as the distribution of local standard deviations in both the foreground and background regions for robust target representation. Appearance learning is then cast as an adaptive Kalman filtering problem where the process and measurement noise variances are both unknown. We formulate this problem using both covariance matching and, for the first time in a visual tracking application, the recent autocovariance least-squares (ALS) method. Although convergence of the ALS algorithm is guaranteed only for the case of globally wide sense stationary process and measurement noises, we demonstrate for the first time that the technique can often be applied with great effectiveness under the much weaker assumption of piecewise stationarity. The performance advantages of the ALS method relative to the classical covariance matching are illustrated by means of simulated stationary and nonstationary systems. Against real data, our results show that the ALS-based algorithm outperforms the covariance matching as well as the traditional histogram similarity-based methods, achieving sub-pixel tracking accuracy against the well-known AMCOM closure sequences and the recent SENSIAC automatic target recognition dataset.

Noise Adaptive LDPC Decoding Using Particle Filtering

Belief propagation (BP) is a powerful algorithm to decode low-density parity check (LDPC) codes over additive white Gaussian noise (AWGN) channels. However, the traditional BP algorithm cannot adapt efficiently to the statistical change of SNR in an AWGN channel. This paper proposes an adaptive scheme that incorporates a particle filtering (PF) algorithm into the BP based LDPC decoding process. The proposed scheme is capable to perform online estimation of time-varying SNR at the bit-level and enhance the BP decoding performance simultaneously.

Braille document recognition using Belief Propagation

There is a significant need for a system to recognize Braille documents in order to preserve them and make them available to a larger group of visually impaired people. A new system for Braille document recognition is proposed. We introduce a highly-adaptive Braille documents parameters estimation method to automatically determine the rotation angle, indentations, and spacing in both vertical and horizontal orientation. The key element in determining the rotation angle of the images is based on Radon transform. Also we introduce the method of Braille translation using Belief Propagation on the assumption that the Braille document is a Hidden Markov Model. We demonstrate the effectiveness of rotation angle correction as well as the accuracy of indentation and spacing in both orientations. We also prove that the translation algorithm is efficient and robust to errors made in the dot detection. The proposed method may be used for further applications.

Theory and Implementation of a Computationally Efficient Decimation Filter for Power-Aware Embedded Systems

As analog-to-digital converters become faster, this will allow them to become closer to their intended sensor. This will foster an environment that will continue to allow a paradigm shift in which digital systems replace analog ones, thus mitigating many nonideal effects, lowering costs, and providing more compact computational platforms. In parallel with this trend, the importance of decimation filters will continue to expand, as the high-speed data will need to be downsampled prior to ingestion by a decision-making element, such as a digital signal processor running constant false alarm rate (CFAR) algorithms, neural networks, and the like. Ideally, these decimation filters should have as much stopband attenuation as possible and should not be hindered by timing bottlenecks. However, on a fixed-point processor, like a field-programmable gate array (FPGA), finite word-length effects are in opposition to this goal. To break this nexus, this paper employs a revolutionary integerization technique based on multidimensional continued fractions strategically coupled with an efficient multiplierless architecture design strategy. Multidimensional continued fractions have been known within the mathematical community for some time, which include the popular Furtwangler algorithm and the ordered Jacobi-Perron algorithm, but have been left unexplored in the engineering community until recently. Simultaneous rational representations (SRRs) are another member of the multidimensional continued fraction family and are employed here to create fixed integer transforms with computationally optimal representations. In addition, this paper also focuses on hardware implementations in low-cost FPGAs or application-specific integrated circuits, which benefit from multiplierless implementation to save hardware real estate. From a computational perspective, carefully choosing how to represent the coefficients of a transform may have dramatic effects on how many operations are consumed to implement it. In a low-order example, the number seven may be expressed as 23-20 or 22 +21+20. This concept coupled with SRRs is explored in this paper to yield low-power high-speed implementations for embedded systems

Target Tracking using a Particle Filter Based on the Projection Method

We present a new particle filter (PF) algorithm, which uses a mathematical tool known as Galerkins projection method to generate the proposal distribution. By definition, Galerkins method is a numerical approach to approximate the solution of a partial differential equation. By leveraging this method with L2 theory and the FFT, this new proposal is fundamentally different to various local linearization or Kalman filter based proposals. We apply this algorithm to a bearings-only tracking problem. As shown in the theory and indicated by our simulations, this proposal renders more support from the true posterior distribution, thereby significantly enhances the estimation accuracy compared to standard bootstrap filters. In addition, because of this improved proposal distribution, the new particle filter can achieve a given level of performance with less sample size.