Olov Rosén

Speedup and tracking accuracy evaluation of parallel particle filter algorithms implemented on a multicore architecture

Four different parallel particle filters such as globally distributed particle filter (GDPF), resampling with proportional allocation filter (RPA), resampling with non-proportional allocation filter (RNA) and the Gaussian particle filter (GPF), are studied in terms of speedup and tracking accuracy in a bearings-only tracking problem. The filters are implemented on a shared memory multicore computer, where the speedup is measured using up to eight cores. The tracking accuracy is studied in a simulated BOT application where the GPF exhibits best tracking accuracy, and RNA, RPA and GDPF give tracking accuracy comparable to the sequential particle filter. Both GPF and RNA appear to be capable of achieving linear speedup in the number of cores used, while RPA shows somewhat less encouraging speedup and GDPF is found to have a speedup limited to about 3 times.

Efficient Parallel Implementation of State Estimation Algorithms on Multicore Platforms

For many applications in signal processing and control it is crucial that estimates of the state vector in a dynamic system can be obtained in real time. This poses the problem of producing algorithms that are fast enough to enable online execution. In this article, it is investigated how two of the most popular and powerful state estimation algorithms, the Kalman filter and the particle filter, can be efficiently implemented in parallel on a multicore architecture. The proposed parallel implementations are analyzed in terms of hardware requirements, such as memory bandwidth and available cache memory, to provide the desired speedup. The algorithms are exemplified by and evaluated in an adaptive filtering and a bearings-only tracking application. In the cases when original algorithms have been modified for parallelization, the accuracy of the estimates obtained is evaluated in comparison with that of the sequential algorithm. It is found that linear speedup, in the number of cores used, can indeed be achieved without loss of accuracy, for both state estimation algorithms.

Nonlinear estimation of a parsimonious Wiener model for the neuromuscular blockade in closed-loop anesthesia

Nonlinear estimation of a parsimonious Wiener model for the neuromuscular blockade in closed-loop anesthesia

An on-line algorithm for anomaly detection in trajectory data

An algorithm for anomaly detection in trajectory data is presented. The algorithm has an intrinsic capability of handling spatial and temporal data shifts as well as dealing with trajectories of unequal lengths and, possibly, non-uniformly sampled in time. Further, it has low computational complexity and can be used in an on-line setting. The main idea of the algorithm is to extract a mean path that is “normal” for the monitored route, and with respect to the mean path, calculate the anomaly score of an acquired trajectory by means of a statistical test. The algorithm is evaluated for a simulated test scenario, where it finds all anomalous trajectories while raising no false alarms. A test on a real data set, containing trajectories of freight ships traveling through the English Channel, also proves the algorithm to perform well.

Efficient parallel implementation of a Kalman filter for single output systems on multicore computational platforms

Parallelization and cache memory bandwidth demand of a Kalman filter for single output systems on multicore computers are investigated and exemplified by an adaptive filtering application. By breaking the data dependencies through a re-organization of calculations, an almost completely parallel algorithm is obtained. Analysis of the resulting algorithm brings about an estimate of the memory bandwidth necessary for a linear in the number of cores speedup. An evaluation of the parallel algorithm on two different shared-memory multicore architectures has been performed. It is found that linear speedup in the number of used cores can indeed be achieved provided a sufficient memory bandwidth is offered by the hardware.

Parallel recursive estimation using Monte Carlo and orthogonal series expansions

Parallelizability of an algorithm is nowadays a highly desirable property as computer hardware is becoming increasingly parallel. In this paper, a formulation of the particle filtering algorithm, suitable for parallel or distributed computing, is proposed. From the particle set, a series expansion is fitted to the posterior probability density function. The global information provided by the particles can in this way be expressed by a few informative coefficients that can be efficiently communicated between the local processing units. Experiments on a shared-memory multicore processor using up to eight cores show that a linear speedup in the number of used cores is achieved.

Parametric and Nonparametric Analysis of Eye-Tracking Data by Anomaly Detection

An approach to smooth pursuit eye movements analysis by means of stochastic anomaly detection is presented and applied to the problem of distinguishing between patients diagnosed with Parkinsons disease and normal controls. Both parametric Wiener model-based techniques and nonparametric modeling utilizing a description of the involved probability density functions in orthonormal bases are considered. The necessity of proper visual stimuli design for the accuracy of mathematical modeling is highlighted and a formal method for producing such stimuli is suggested. The efficacy of the approach is demonstrated on experimental data collected by means of a commercial video-based eye tracker.

The Recursive Bayesian Estimation Problem via Orthogonal Expansions: an Error Bound

Abstract When solving the non linear non Gaussian filtering problem via orthognal series expansions the involved probability density functions are approximated with truncated series expansions. Inevitable the truncation introduces an error. In this paper an upper bound on the 1-norm of the approximation error in the probability density function of the state vector conditional on the system output measurements, due to the truncation, is derived and numerically evaluated in a simulation example. The bound quantifies the proximity of the obtained approximate solution to the true one. To explore the choice of orthonormal basis as a degree of freedom in the proposed method, a comparison between the Fourier and Legendre bases in a bearings-only tracking problem is performed.

A LEGO-based mobile robotic platform for evaluation of parallel control and estimation algorithms

An inexpensive robotic system intended for educational use in parallel algorithms for embedded control and signal processing is described. The hardware platform is comprised of a state-of-the-art multi-core system in a wireless network with several mobile LEGO robots that collect data from their environment. The setup covers a broad range of real-time cooperative and parallel problems arising in sensor networks, robotics, surveillance and high-performance embedded applications. As an illustration, a bearings-only tracking problem, estimating both mobile robots positions and the position of a non-cooperating target by using parallel particle filtering, is solved on the proposed platform. In order to improve the estimation accuracy and to adjust to changes in the environment and movements of the target, a controller positioning the mobile robots is utilized.

Parallelization of the Kalman filter for banded systems on multicore computational platforms

Parallel implementation of the Kalman filter (KF), with emphasis on multicore architecture implementation is investigated. It is shown that KF provides enough parallelizm to achieve linear speedup in the number of cores used. A computationally efficient implementation that exhibit good scalability properties is presented. The implementation is based on the assumption of a banded system matrix. Both time-varying and invariant systems can be generally transformed to a realization with a banded system matrix. An analysis of the algorithm that provides guidelines to the choice of implementation hardware to meet a desired performance is also given.