Showing papers by "Simon Maskell published in 2005"

Poisson models for extended target and group tracking

Abstract: It is common practice to represent a target group (or an extended target) as set of point sources and attempt to formulate a tracking filter by constructing possible assignments between measurements and the sources. We suggest an alternative approach that produces a measurement model (likelihood) in terms of the spatial density of measurements over the sensor observation region. In particular, the measurements are modelled as a Poisson process with a spatially dependent rate parameter. This representation allows us to model extended targets as an intensity distribution rather than a set of points and, for a target formation, it gives the option of modelling part of the group as a spatial distribution of target density. Furthermore, as a direct consequence of the Poisson model, the measurement likelihood may be evaluated without constructing explicit association hypotheses. This considerably simplifies the filter and gives a substantial computational saving in a particle filter implementation. The Poisson target-measurement model will be described and its relationship to other filters will be discussed. Illustrative simulation examples will be presented.

Recursive track-before-detect with target amplitude fluctuations

Abstract: A particle-based track-before-detect filtering algorithm is presented. This algorithm incorporates the Swerling family of target amplitude fluctuation models in order to capture the effect of radar cross-section changes that a target would present to a sensor over time. The filter is designed with an existence variable, to determine the presence of a target in the data, and an efficient method of incorporating this variable in a particle filter scheme is developed. Results of the algorithm on simulated data show a significant gain in detection performance through accurately modelling the target amplitude fluctuations.

A unifying framework for multi-target tracking and existence

Abstract: Most target tracking approaches either assume that the number of targets is constant throughout the time horizon of interest, or that information about target existence (birth and death) is provided by some external source. Here we show how target existence can be integrated within the tracking framework in a rigorous way. The notion of existence is not new, and has been considered before in e.g. [D. Musicki et al., (1994), (2002)]. We provide here a general probabilistic treatment that impacts as little as possible on existing tracking algorithms so that legacy tracking software (and more generally target tracking architectures) can be reused. We first show how the notion of existence can be incorporated into a single target tracking framework (retaining algorithmic invariance). To place the probabilistic recursions into context we relate this single target tracking architecture to the probabilistic data association filter. We then extend the single target results to incorporate existence for multi-target tracking and relate this to an importance sampling implementation of the joint probabilistic data association (JPDA) framework. The treatment presented is entirely general and so facilitates implementation with Kalman filters, extended/unscented Kalman filters, particle filters, etc, i.e. the approach developed is invariant to the filtering and data association mechanisms used, and therein lies the novelty. We apply the proposed framework to the difficult problem of tracking football players in video sequences, where we adopt a mixture Kalman filter implementation.

Online sensor registration

Abstract: In a multi-sensor scenario, accurate data fusion is best achieved by processing the measurements from all the sensors at a fusion node to produce tracks. However, inaccuracies in the position and/or rotation of the sensor can lead to "ghost" tracks, particularly when the sensors are not co-located. This paper presents a framework which models the uncertainty over the sensors' registration parameter (e.g. position and rotation) and discloses an unscented implementation technique (other methods based on particle filters can be accommodated within our framework), where each sensor self-localises using targets of opportunity. The aim is to solve the sensor registration problem whilst adding minimal overhead to an existing tracker, which is facilitated by making the standard assumption that the state of the joint target factorises over the individual targets

Tracking using a radar and a problem specific proposal distribution in a particle filter

Abstract: Particle filters are the state-of-the-art solution to difficult tracking problems. They describe the uncertainty associated with a track using the diversity of a set of samples. These samples are the particles and they each represent a hypothesis for the state of the target. The crucial step in the efficient application of particle filters to specific problems is the design of the proposal distribution. This proposal distribution defines how the particles are propagated from one time step to the next. If the proposal is not well designed then one typically needs a huge number of particles to achieve good performance. Therefore the paper focuses on the design of a proposal distribution that is well suited to the specific problem being considered; this work is motivated by the need to use particle filters to track using multiple radars observing multiple closely spaced targets at high range. The targets can be resolved in radar co-ordinates, but not easily tracked in Cartesian co-ordinates. In the general case, it often happens that the proposal needs to interpolate between both the information derived using the dynamic model for the target behaviour and the information inherent in the measurement. Schemes to conduct this interpolation are often based on extended or unscented Kalman filters. As has been known for some time, when using such filters in nonlinear environments such as when tracking using radars, a pertinent choice of co-ordinate frame can improve performance. Based on this idea, a proposal distribution is described that uses an extended Kalman filter in radar co-ordinates. This results in a skewed proposal in Cartesian co-ordinates that is shown, in this specific problem, to improve on the performance possible using a particle filter with other choices of proposal distribution.

Bayesian visual tracking with existence process

Abstract: Most object tracking approaches either assume that the number of objects is constant, or that information about object existence is provided by some external source. Here, we show how object existence can be rigorously integrated within the Bayesian single and multiple object tracking framework. We provide a general treatment that impacts as little as possible on existing tracking algorithms, so that software can be reused, and that allows implementation with Kalman filters, extended Kalman filters, particle filters, etc. We apply the proposed framework to colour-based tracking of multiple objects.

Joint target tracking and identification. Part II. Shape video computing

Abstract: For Pt. I see ibid., vol.1 p.256-299, (2005). This paper describes an application of sequential Monte Carlo model-based approaches to perform joint target tracking and identification. While a geometric shape is moving inside the field of view of a CCD camera, alternatively getting closer and moving away while rotating, the data processing system is confronted to challenging tasks: track the moving shape in real 3D space, i.e. estimate its position and orientation, and at the same time dynamically estimate its dimensions and, if required, identify it. The system is based on class-specific Bayesian filters. More originally, the issue of the fixed hyper-parameter estimation, here the geometric shape dimensions, is solved by combining two different techniques. The first one consists of Markov chain Monte Carlo moves that rescale both the trajectory and the shape; it benefits from an efficient statistic which summaries the trajectory with regard to moves. The second one is an artificial deformation diffusion of the shape.

Joint target tracking and identification-Part I: sequential Monte Carlo model-based approaches

Abstract: This paper deals with model-based approaches for joint target tracking and identification In a Bayesian framework, parametric state-space model classes are introduced as a generalization of the widespread state-space models In addition to the dynamic state, they include a hyper-parameter, which takes into account target features or behaviors For such model classes, sequential Monte Carlo approaches, also known as particle filtering, provide a powerful tool to perform sequentially on-line estimation and model selection The paper focuses on the ergodicity concern of fixed hyper-parameter estimation and model selection Indeed, the infinite memory of such a system may lead to the particle filter degeneracy or divergence It reviews various methods to solve this problem, from the common and basic trick of adding an artificial noise to more complex methods, such as the introduction of reversible jump Markov chain Monte Carlo moves

Dynamic sensor coalition formation to assist the distributed tracking of targets: application to wide-area surveillance

Abstract: The protection of infrastructure and facilities within the UK is of prime importance in the current environment where terrorist threats are present. Surveillance of large areas within such facilities is a complex, manpower intensive and demanding task. To reduce the demands on manpower, new systems will need to be developed that use a mixed sensor suite associated with access to databases containing historical data and known threats. This requires fusion of mixed type data from disparate sources. The methods used for the fusion process, and the location of the fusion process, will be dependent on the data, sensor or database. The communication requirements will also be of paramount importance within the monitoring network. As computers increase in performance and reduce in cost and power consumption, there is a growing trend for more processing to be carried out locally. This raises issues of compatibility, timeliness, global awareness of the situation and distributed versus centralised control of the system. This paper presents a generic solution to the wide-area surveillance problem through the application and combination of Covariance In∞ation (a distributed fusion mathematical c ∞ 2005 framework that circumvents problems with data incest) with agent-based technologies (allowing the dynamic formation of sensor coalitions) to track, and potentially risk assess, targets within the region of interest. A discussion will be provided into the distributed detection and tracking of an intruding vehicle at a commercial airport to place the seemingly abstract technology into context.