Typical surveillance systems employ decision- or feature-level fusion approaches to integrate heterogeneous sensor data, which are sub-optimal and incur information loss. In this paper, we investigate data-level heterogeneous sensor fusion. Since the sensors monitor the common targets of interest, whose states can be determined by only a few parameters, it is reasonable to assume that the measurement domain has a low intrinsic dimensionality. For heterogeneous sensor data, we develop a joint-sparse data-level fusion (JSDLF) approach based on the emerging joint sparse signal recovery techniques by discretizing the target state space. This approach is applied to fuse signals from multiple distributed radio frequency (RF) signal sensors and a video camera for joint target detection and state estimation. The JSDLF approach is data-driven and requires minimum prior information, since there is no need to know the time-varying RF signal amplitudes, or the image intensity of the targets. It can handle non-linearity in the sensor data due to state space discretization and the use of frequency/pixel selection matrices. Furthermore, for a multi-target case with J targets, the JSDLF approach only requires discretization in a single-target state space, instead of discretization in a J-target state space, as in the case of the generalized likelihood ratio test (GLRT) or the maximum likelihood estimator (MLE). Numerical examples are provided to demonstrate that the proposed JSDLF approach achieves excellent performance with near real-time accurate target position and velocity estimates.
In a distributed radar track fusion system, it is desired to limit the communication rate between the sensors and the central node to only the most relevant information available. One way to do this is to use some metric that judges quantity of new information available, in comparison to that which has already been provided. The J-Divergence is a symmetric metric, derived from the Kullback-Liebler divergence, which performs a comparison of the statistical distance between two probability distributions. For the comparison between new and old data, a large J-Divergence can represent the existence of new information, while a small J-Divergence represents the lack of new information. Previous work included an application where the J-Divergence was used to limit data for scenarios in which the primary state estimator was an Extended Kalman Filter and used only Gaussian approximations at the local sensors. This paper expands the range of estimators to particle filters in order to account for situations where censoring is desired to be applied to non-linear/non-Gaussian environments. A derivation of the J-Divergence between probability density functions (PDFs) which are approximated by particles is provided for use in a non-feedback fusion case. An example application is given involving a 2D radar tracking scenario using the J-Divergences of a particle filter with the Gaussian approximation and a particle filter with the approximated discrete prior/posterior PDFs.
System state estimation in the presence of an adversary that injects false information into sensor readings has attracted much attention in wide application areas, such as target tracking with compromised sensors, secure monitoring of dynamic electric power systems, secure driverless cars, and radar tracking and detection in the presence of jammers. From a malicious adversary’s perspective, the optimal strategy for attacking a multi-sensor dynamic system over sensors and over time is investigated. It is assumed that the system defender can perfectly detect the attacks and identify and remove sensor data once they are corrupted by false information injected by the adversary. With this in mind, the adversary’s goal is to maximize the covariance matrix of the system state estimate by the end of attack period under a sparse attack constraint such that the adversary can only attack the system a few times over time and over sensors. The sparsity assumption is due to the adversary’s limited resources and his/her intention to reduce the chance of being detected by the system defender. This becomes an integer programming problem and its optimal solution, the exhaustive search, is intractable with a prohibitive complexity, especially for a system with a large number of sensors and over a large number of time steps. Several suboptimal solutions, such as those based on greedy search and dynamic programming are proposed to find the attack strategies. Examples and numerical results are provided in order to illustrate the effectiveness and the reduced computational complexities of the proposed attack strategies.
Conference Committee Involvement (3)
Signal Processing, Sensor/Information Fusion, and Target Recognition XXXIV
14 April 2025 | Orlando, Florida, United States
Signal Processing, Sensor/Information Fusion, and Target Recognition XXXIII
22 April 2024 | National Harbor, Maryland, United States
Signal Processing, Sensor/Information Fusion, and Target Recognition XXXII
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.