Buch, Englisch, Band 7, 278 Seiten, PB, Format (B × H): 1480 mm x 2100 mm, Gewicht: 430 g
Reihe: Fraunhofer Series Advances in Sensor Data and Information Fusion
Buch, Englisch, Band 7, 278 Seiten, PB, Format (B × H): 1480 mm x 2100 mm, Gewicht: 430 g
Reihe: Fraunhofer Series Advances in Sensor Data and Information Fusion
ISBN: 978-3-89863-262-1
Verlag: GCA
Abstract
The production of a detailed situation picture of a generally complex and dynamically evolving scenario is of great importance in many areas, e.g., border protection, wide-area surveillance of remote areas or support of disaster management and emergency services. Such a situation picture contains condensed information about the objects of interest, in particular, their existence, number, locations and motion histories. Based on these constituents, further conclusions can then be drawn about, e.g., the behavior, individual properties, identities, interrelations and communication channels among different objects as well as a detailed traffic analysis and anomaly detection. One essential building block for the production of such a detailed situation picture are tracks, i.e., state estimate sequences, of moving objects, based on detections from a single or multiple sensors.
The topic of this thesis is the improvement of state-of-the-art Bayesian tracking filters specialized to the domain of ground moving objects to obtain high-quality track information in terms of track precision and track stability. It is assumed that a given scenery on the ground containing well-separated or even groups of closely-spaced objects is observed by an airborne radar system which is well suited for this task due to its wide-area surveillance, day & night operation as well as real-time processing capabilities.
The tracking of ground moving objects based on airborne radar measurements generally faces several challenges which strongly deteriorate the performance of standard tracking filters. The major challenges are imprecise measurements and missed detections, a strong false alarm background, closely-spaced targets, technical and terrain obscuration as well as complex target motion. To counterbalance the strong performance degradation, it is therefore necessary to introduce advanced methods for exploiting additional sources of information in a tracking system, i.e., to fuse the already available sensor data with additional sensor, attribute and context information within an advanced tracking filter.
In this thesis, different classes of information are considered and used as extensions of standard tracking algorithms. In particular, contributions are made to the areas of blind zone knowledge, road network information and signal strength measurement processing. Each of these classes is briefly commented in the following: The presence of a sensor's blind zone generally leads to sequences of missed detections whenever a target under track is masked by such a blind zone. A prominent example is the blind zone in Doppler that affects the detection of those objects whose radial velocity component relative to the sensor is too low so that a discrimination from the general clutter background is not possible. Another example is the blind zone in the range domain, caused by the switching between transmit and receive mode if only a single antenna system is employed for both tasks. The second considered source of information is road-map data: Most ground moving objects travel along predefined paths, i.e., roads, provided by the given infrastructure of a certain area. In addition, digitized road-map data is a commonly available source of information. It is therefore reasonable to consider this context information as additional building block for the composition of an advanced Bayesian tracking filter. Finally, the signal strength of a radar detection is a standard output of modern radar systems. In this thesis, the signal strength information is used to sequentially estimate a target's mean RCS along with the kinematic state vector, assuming that the fluctuations of the target backscattering cross section which result in fluctuations of the signal strength measurement can be described by statistical fluctuation models.
Based on a systematic analysis of simulations as well as real data, it is demonstrated that the exploitation of each developed filter extension alone leads to a gain in tracking performance in specific situations of a realistic ground scenario: Exploiting extended blind zone knowledge counterbalances track deteriorations due to blind zone masking. The incorporation of road network information leads to a strong increase in track precision. And utilizing signal strength measurements for the estimation of a target's mean backscattering cross section yields valuable information for the discrimination of closely-spaced targets. But it is demonstrated that only a sophisticated combination of these complementary sources of information yields a tremendous gain in both track precision and track stability in a wide range of situations that occur during a realistic scenario comprising ground moving objects.
The key contributions of this thesis can be summarized as follows:
• Increased blind zone knowledge is made available by developing a method to determine the width of the Doppler blind zone for arbitrary radar sensor configurations and antenna orientations and by also introducing the handling of blind zones in the range domain.
• A road network processing scheme based on adaptive local roads is developed where ambiguities arising at crossings and junctions are resolved over time using a multiple model approach.
• A generalized estimation scheme for the mean backscattering cross section is developed that is suitable for ground targets in a cluttered environment.
• A general integrated Bayesian update scheme is derived incorporating all previous individual filter extensions. This general scheme is then implemented into three different standard target tracking algorithms.
• The capability of the highly augmented Bayesian tracking filters is then evaluated systematically based on single- and multiple target scenarios, yielding a tremendous performance gain in a wide range of situations in both track precision as well as track stability compared to standard or less augmented filter variants.
