doc. Ing. Kamil Dedecius, Ph.D.

Článek

10.1109/LSP.2025.3642058

Katedra aplikované matematiky

The letter investigates the problem of distributed multitarget tracking with a network of sensors with limited and partially overlapping or non-overlapping fields of view. The information processing is based on information diffusion, where each sensor can communicate only with its adjacent neighbors. The communication comprises an adaptation phase suited for the exchange of measurements, followed by a combination phase where the estimates are shared and fused via arithmetic average rule. Each phase is performed only once at each discrete time step, thus effectively reducing computational, memory, and communication overheads. An important part of the solution is the self-referencing mechanism, allowing the incorporation of only those neighbors' information that aligns with local estimates or enhances them. The simulation example demonstrates improved localization performance and resilience to misdetections.

Stať ve sborníku

Katedra aplikované matematiky

The rapid development of smart systems such as autonomous vehicles (AVs) and unmanned aerial systems (UASs) imposes substantial requirements on the real-time monitoring of surrounding moving objects. In this paper, we propose an extension of the popular multi-object filter, namely the Joint Integrated Probabilistic Data Association (JIPDA) filter, for networks of agents (e.g., AVs and UASs) equipped with short-range communication capabilities. By sharing observations acquired within their respective fields of view, agents can significantly enhance object tracking performance in cluttered environments. The communication requirements of the proposed solution are relatively modest and theoretically suitable for the considered applications.

Stať ve sborníku

10.23919/EUSIPCO63174.2024.10715094

Katedra aplikované matematiky

The paper introduces a novel collaborative method for multi-target tracking in cluttered environments. Based on the information diffusion framework, the proposed approach involves agents sharing observations and/or posterior estimates with neighboring agents within one hop distance. No intermediate iterations are employed to achieve global agreement, as in consensus algorithms, nor is there a single point of failure, as in algorithms with a fusion center. The tracking method is based on the joint integrated probabilistic data association (JIPDA) filter, which is adapted to accurately incorporate neighbors' observations and optimally fuse available estimates. The ultimate goal is to improve the tracking performance by overcoming the intrinsic uncertainty of target-measurement association and object detection. The method is computationally cheap and does not require any numerical optimization. Simulation results clearly demonstrate the performance of the proposed method.

Článek

10.1109/LSP.2020.2987723

Fakulta informačních technologií

The Poisson regression is a popular model for positive integer random variables determined by known explanatory variables. This letter studies the problem of its collaborative Bayesian sequential estimation under potentially slowly time-varying regression coefficients. We assume networks where agents share their information about the inferred quantities with adjacent neighbors in order to improve the overall estimation performance. The communication strategy is the information diffusion, i.e., only one information exchange per time instant is allowed.