Pub Date : 2021-11-01DOI: 10.23919/fusion49465.2021.9627008
J. J. Steckenrider, Brock Crawford, Penny Zheng
This paper proposes a framework for fusing information coming from an independent inertial measurement unit (IMU) and global positioning system (GPS) to deliver robust estimation of human gait. Because these two sensors provide very different kinds of data at different scales and frequencies, a novel approach which fuses global trajectory estimates and back-propagates this information to correct step vectors is put forth here. In several high-fidelity simulations, the proposed technique is shown to improve step estimation error up to 40% in comparison with an IMU-only approach. This work has implications for not only in-the-field biomechanics research, but also cooperative field robotic systems where it may be critical to accurately monitor a person’s position and state in real-time.
{"title":"GPS and IMU Fusion for Human Gait Estimation","authors":"J. J. Steckenrider, Brock Crawford, Penny Zheng","doi":"10.23919/fusion49465.2021.9627008","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9627008","url":null,"abstract":"This paper proposes a framework for fusing information coming from an independent inertial measurement unit (IMU) and global positioning system (GPS) to deliver robust estimation of human gait. Because these two sensors provide very different kinds of data at different scales and frequencies, a novel approach which fuses global trajectory estimates and back-propagates this information to correct step vectors is put forth here. In several high-fidelity simulations, the proposed technique is shown to improve step estimation error up to 40% in comparison with an IMU-only approach. This work has implications for not only in-the-field biomechanics research, but also cooperative field robotic systems where it may be critical to accurately monitor a person’s position and state in real-time.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"51 3 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116542861","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.23919/fusion49465.2021.9627009
David Schwab, Sean M. O’Rourke, Breton L. Minnehan
Trajectory forecasting is vital to target tracking, autonomous decision making, and other fields critical to the future of autonomous systems. Tracking algorithms, such as the Kalman Filter, require accurate motion models in order to forecast target trajectories and update state estimates given observation data. Unfortunately, accurate motion models are not always easily de- fined. Of particular interest is forecasting in systems with complex agent-to-agent and agent-to-scene interactions, which are often best represented as a multimodal distribution. Various network architectures tackle this multimodal problem in different ways, but the method used in this work is a mixture density network. The network architecture examined in this work, LSTM2MDN, builds off previous research in combining the renowned long- short term memory (LSTM) network with a mixture density network (MDN) in order to develop accurate distributions for output trajectories.
{"title":"Combining LSTM and MDN Networks for traffic forecasting using the Argoverse Dataset","authors":"David Schwab, Sean M. O’Rourke, Breton L. Minnehan","doi":"10.23919/fusion49465.2021.9627009","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9627009","url":null,"abstract":"Trajectory forecasting is vital to target tracking, autonomous decision making, and other fields critical to the future of autonomous systems. Tracking algorithms, such as the Kalman Filter, require accurate motion models in order to forecast target trajectories and update state estimates given observation data. Unfortunately, accurate motion models are not always easily de- fined. Of particular interest is forecasting in systems with complex agent-to-agent and agent-to-scene interactions, which are often best represented as a multimodal distribution. Various network architectures tackle this multimodal problem in different ways, but the method used in this work is a mixture density network. The network architecture examined in this work, LSTM2MDN, builds off previous research in combining the renowned long- short term memory (LSTM) network with a mixture density network (MDN) in order to develop accurate distributions for output trajectories.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"100 6 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125299870","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.23919/fusion49465.2021.9626888
S. Coraluppi, C. Carthel, R. Prengaman
Sensors with poor bearing resolution pose a significant challenge for multi-target tracking, as cross-range error becomes very large at long ranges. While multi-sensor fusion provides benefit towards higher-precision tracking, there are two key difficulties to confront. The first is to address measurement association ambiguities, which we address via advanced multiple-hypothesis tracking. The second is to perform robust track initialization and filtering, which we achieve via a two-point filter initialization approach followed by (sequential) extended Kalman filtering. In the specific context of active sonar tracking, the impact of finite sound speed poses an additional challenge. Addressing this requires a generalized MHT solution that accounts for measurement-specific time stamps and allows for out-of-sequence measurement processing. The enhancements discussed in this paper yield a robust capability for wide-area multistatic sonar tracking.
{"title":"Wide-Area Multistatic Sonar Tracking","authors":"S. Coraluppi, C. Carthel, R. Prengaman","doi":"10.23919/fusion49465.2021.9626888","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626888","url":null,"abstract":"Sensors with poor bearing resolution pose a significant challenge for multi-target tracking, as cross-range error becomes very large at long ranges. While multi-sensor fusion provides benefit towards higher-precision tracking, there are two key difficulties to confront. The first is to address measurement association ambiguities, which we address via advanced multiple-hypothesis tracking. The second is to perform robust track initialization and filtering, which we achieve via a two-point filter initialization approach followed by (sequential) extended Kalman filtering. In the specific context of active sonar tracking, the impact of finite sound speed poses an additional challenge. Addressing this requires a generalized MHT solution that accounts for measurement-specific time stamps and allows for out-of-sequence measurement processing. The enhancements discussed in this paper yield a robust capability for wide-area multistatic sonar tracking.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"137 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122928803","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.23919/fusion49465.2021.9626975
Daniel Frisch, U. Hanebeck
We propose a simple and efficient method to obtain unweighted deterministic samples of the multivariate Gaussian density. It allows to place a large number of homogeneously placed samples even in high-dimensional spaces. There is a demand for large high-quality sample sets in many nonlinear filters. The Smart Sampling Kalman Filter (S2KF), for example, uses many samples and is an extension of the Unscented Kalman Filter (UKF) that is limited due to its small sample set. Generalized Fibonacci grids have the property that if stretched or compressed along certain directions, the grid points keep approximately equal distances to all their neighbors. This can be exploited to easily obtain deterministic samples of arbitrary Gaussians. As the computational effort to generate these anisotropically scalable point sets is low, generalized Fibonacci grid sampling appears to be a great new source of large sample sets in high-quality state estimation.
{"title":"Deterministic Gaussian Sampling With Generalized Fibonacci Grids","authors":"Daniel Frisch, U. Hanebeck","doi":"10.23919/fusion49465.2021.9626975","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626975","url":null,"abstract":"We propose a simple and efficient method to obtain unweighted deterministic samples of the multivariate Gaussian density. It allows to place a large number of homogeneously placed samples even in high-dimensional spaces. There is a demand for large high-quality sample sets in many nonlinear filters. The Smart Sampling Kalman Filter (S2KF), for example, uses many samples and is an extension of the Unscented Kalman Filter (UKF) that is limited due to its small sample set. Generalized Fibonacci grids have the property that if stretched or compressed along certain directions, the grid points keep approximately equal distances to all their neighbors. This can be exploited to easily obtain deterministic samples of arbitrary Gaussians. As the computational effort to generate these anisotropically scalable point sets is low, generalized Fibonacci grid sampling appears to be a great new source of large sample sets in high-quality state estimation.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"14 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127825794","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.23919/fusion49465.2021.9626875
M. Trivedi, J. V. Wyk
Complex systems are in place for the localization and tracking of High-speed Trains. These methods tend to perform poorly under certain conditions. Localization using 5G infrastructure has been considered as an alternative solution for the positioning of trains in previous studies. However, these studies only consider localization using Time Difference of Arrival measurements or using Time of Arrival and Angle of Departure measurements. In this paper an alternate compressed sensing based 5G localization method is considered for this problem. The proposed algorithm, paired with an Extended Kalman Filter, is implemented and tested on a 3GPP specified high s peed train scenario. Sub-meter localization accuracy was achieved using 4-6 Remote-Radio-Heads, while an accuracy of 0.34 m with 95% availability is achieved when using 2 Remote-Radio-Heads. The achieved performance meets 3GPP specified requirement for machine control and transportation even when using 2 Remote-Radio-Heads.
用于高速列车定位和跟踪的复杂系统已经到位。这些方法在某些条件下往往表现不佳。在之前的研究中,利用5G基础设施进行定位被认为是列车定位的另一种解决方案。然而,这些研究仅使用到达时差测量或使用到达时间和出发角测量来考虑定位。本文考虑了一种基于压缩感知的5G定位方法。该算法与扩展卡尔曼滤波相结合,在3GPP高速列车场景中进行了实现和测试。使用4-6个remote - radio - head可实现亚米级定位精度,而使用2个remote - radio - head可实现0.34 m的精度和95%的可用性。即使使用2个remote - radio - head,也能满足3GPP对机器控制和运输的要求。
{"title":"Localization and Tracking of High-speed Trains Using Compressed Sensing Based 5G Localization Algorithms","authors":"M. Trivedi, J. V. Wyk","doi":"10.23919/fusion49465.2021.9626875","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626875","url":null,"abstract":"Complex systems are in place for the localization and tracking of High-speed Trains. These methods tend to perform poorly under certain conditions. Localization using 5G infrastructure has been considered as an alternative solution for the positioning of trains in previous studies. However, these studies only consider localization using Time Difference of Arrival measurements or using Time of Arrival and Angle of Departure measurements. In this paper an alternate compressed sensing based 5G localization method is considered for this problem. The proposed algorithm, paired with an Extended Kalman Filter, is implemented and tested on a 3GPP specified high s peed train scenario. Sub-meter localization accuracy was achieved using 4-6 Remote-Radio-Heads, while an accuracy of 0.34 m with 95% availability is achieved when using 2 Remote-Radio-Heads. The achieved performance meets 3GPP specified requirement for machine control and transportation even when using 2 Remote-Radio-Heads.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"193 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114095278","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.23919/fusion49465.2021.9626873
Xingxiang Xie, Yang Wang
In a multi-target tracking (MTT) scenario, the computational cost of usual Poisson multi-Bernoulli mixture (PMBM) filter will rise rapidly as the increasing number of global hypotheses. In order to lower computational cost, this paper presents to apply recycling algorithm to PMBM filter. The proposed method is done by recycling Bernoulli components which are less than a fixed threshold, approximate them as Poisson point process (PPP), thus add the intensity to the undetected PPP intensity. In the numerical experiment, we apply recycling algorithm to PMBM, Poisson multi-Bernoulli (PMB) and multi-Bernoulli mixture (MBM), respectively. The result shows that the Bernoulli recycling algorithm leads to lower computational cost in a simulated scenario.
{"title":"Analysis of recycling performance in Poisson multi-Bernoulli mixture filters","authors":"Xingxiang Xie, Yang Wang","doi":"10.23919/fusion49465.2021.9626873","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626873","url":null,"abstract":"In a multi-target tracking (MTT) scenario, the computational cost of usual Poisson multi-Bernoulli mixture (PMBM) filter will rise rapidly as the increasing number of global hypotheses. In order to lower computational cost, this paper presents to apply recycling algorithm to PMBM filter. The proposed method is done by recycling Bernoulli components which are less than a fixed threshold, approximate them as Poisson point process (PPP), thus add the intensity to the undetected PPP intensity. In the numerical experiment, we apply recycling algorithm to PMBM, Poisson multi-Bernoulli (PMB) and multi-Bernoulli mixture (MBM), respectively. The result shows that the Bernoulli recycling algorithm leads to lower computational cost in a simulated scenario.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"32 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117079198","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.23919/fusion49465.2021.9627034
Ashkan Moradi, Naveen K. D. Venkategowda, S. Talebi, Stefan Werner
Distributed filtering techniques have emerged as the dominant and most prolific class of filters used in modern monitoring and surveillance applications, such as smart grids. As these techniques rely on information sharing among agents, user privacy and information security have become a focus of concern. In this manuscript, a privacy-preserving distributed Kalman filter (PP-DKF) is derived that maintains privacy by decomposing the information into public and private substates, where only a perturbed version of the public substate is shared among neighbors. The derived PP-DKF provides privacy by restricting the amount of information exchanged with state decomposition and conceals private information by injecting a carefully designed perturbation sequence. A thorough analysis is performed to characterize the privacy-accuracy trade-offs involved in the distributed filter, with privacy defined as the mean squared estimation error of the private information at the honest-but-curious agent. The resulting PP-DKF improves the overall filtering performance and privacy of all agents compared to distributed Kalman filters employing contemporary privacy-preserving average consensus techniques. Several simulation examples corroborate the theoretical results.
{"title":"Securing the D istributed Kalman Filter Against Curious Agents","authors":"Ashkan Moradi, Naveen K. D. Venkategowda, S. Talebi, Stefan Werner","doi":"10.23919/fusion49465.2021.9627034","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9627034","url":null,"abstract":"Distributed filtering techniques have emerged as the dominant and most prolific class of filters used in modern monitoring and surveillance applications, such as smart grids. As these techniques rely on information sharing among agents, user privacy and information security have become a focus of concern. In this manuscript, a privacy-preserving distributed Kalman filter (PP-DKF) is derived that maintains privacy by decomposing the information into public and private substates, where only a perturbed version of the public substate is shared among neighbors. The derived PP-DKF provides privacy by restricting the amount of information exchanged with state decomposition and conceals private information by injecting a carefully designed perturbation sequence. A thorough analysis is performed to characterize the privacy-accuracy trade-offs involved in the distributed filter, with privacy defined as the mean squared estimation error of the private information at the honest-but-curious agent. The resulting PP-DKF improves the overall filtering performance and privacy of all agents compared to distributed Kalman filters employing contemporary privacy-preserving average consensus techniques. Several simulation examples corroborate the theoretical results.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"34 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115317325","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.23919/fusion49465.2021.9626840
P. Chauchat, D. Medina, J. Vilà‐Valls, É. Chaumette
Precise navigation solutions are fundamental for new intelligent transportation systems and robotics applications, where attitude also plays an important role. Among the different technologies available, Global Navigation Satellite Systems (GNSS) are the main source of positioning data. In the GNSS context, carrier phase observations are mandatory to obtain precise positioning, and multiple antenna setups must be considered for attitude determination. Position and attitude estimation have been traditionally tackled in a separate manner within the GNSS community, but a recently introduced recursive joint position and attitude (JPA) Kalman filter-like approach has shown the potential benefits of the joint estimation. One of the drawbacks of the original JPA is the assumption of perfect system knowledge, and in particular the baseline distance between antennas, which may not be the case in real-life applications and can lead to a severe performance degradation. The goal of this contribution is to propose a robust filtering approach able to mitigate the impact of a possible GNSS antenna baseline mismatch, exploiting the use of linear constraints. Illustrative results are provided to support the discussion and show the performance improvement, for both GNSS-based attitude-only and JPA estimation.
{"title":"Robust Linearly Constrained Filtering for GNSS Position and Attitude Estimation under Antenna Baseline Mismatch","authors":"P. Chauchat, D. Medina, J. Vilà‐Valls, É. Chaumette","doi":"10.23919/fusion49465.2021.9626840","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626840","url":null,"abstract":"Precise navigation solutions are fundamental for new intelligent transportation systems and robotics applications, where attitude also plays an important role. Among the different technologies available, Global Navigation Satellite Systems (GNSS) are the main source of positioning data. In the GNSS context, carrier phase observations are mandatory to obtain precise positioning, and multiple antenna setups must be considered for attitude determination. Position and attitude estimation have been traditionally tackled in a separate manner within the GNSS community, but a recently introduced recursive joint position and attitude (JPA) Kalman filter-like approach has shown the potential benefits of the joint estimation. One of the drawbacks of the original JPA is the assumption of perfect system knowledge, and in particular the baseline distance between antennas, which may not be the case in real-life applications and can lead to a severe performance degradation. The goal of this contribution is to propose a robust filtering approach able to mitigate the impact of a possible GNSS antenna baseline mismatch, exploiting the use of linear constraints. Illustrative results are provided to support the discussion and show the performance improvement, for both GNSS-based attitude-only and JPA estimation.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"45 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123895496","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.23919/fusion49465.2021.9626872
Vusumuzi Malele, M. E. Letsoalo, M. Mafu
The firm’s financial characteristics affecting the audit fees are determined based on the 2099 firms listed on the Compustat database from 2009-2019. A more comprehensive view of this subject is provided by analyzing fundamental financial, statistical, and market information from thousands of companies worldwide based on the database. The best set of predictor variables are identified using descriptive statistics, correlation matrices, and exploratory data analysis. A regression model is built to test and measure the relationship and significance between these predictor variables and audit fees. Notably, results confirm that the firm financial characteristics ACT, INVT, LCT, AT, EBIT, EBITDA, and CEQ determine audit fees. Furthermore, the audit fees are negatively and significantly related to PIFO, FYEAR, EMP, and GVKEY. Previously, studies focused on determinants such as firm size, status of the audit firm, and corporate complexity. Thus, this work integrates an international financial perspective in the determination of audit fees.
{"title":"Determinants of audit fees: Evidence from Compustat database from 2009-2019","authors":"Vusumuzi Malele, M. E. Letsoalo, M. Mafu","doi":"10.23919/fusion49465.2021.9626872","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626872","url":null,"abstract":"The firm’s financial characteristics affecting the audit fees are determined based on the 2099 firms listed on the Compustat database from 2009-2019. A more comprehensive view of this subject is provided by analyzing fundamental financial, statistical, and market information from thousands of companies worldwide based on the database. The best set of predictor variables are identified using descriptive statistics, correlation matrices, and exploratory data analysis. A regression model is built to test and measure the relationship and significance between these predictor variables and audit fees. Notably, results confirm that the firm financial characteristics ACT, INVT, LCT, AT, EBIT, EBITDA, and CEQ determine audit fees. Furthermore, the audit fees are negatively and significantly related to PIFO, FYEAR, EMP, and GVKEY. Previously, studies focused on determinants such as firm size, status of the audit firm, and corporate complexity. Thus, this work integrates an international financial perspective in the determination of audit fees.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121284234","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.23919/fusion49465.2021.9626863
Xi Li, Yi Liu, Le Yang, L. Mihaylova, Bing Deng
This paper considers the problem of fixed-interval smoothing for Markovian switching systems with multiple linear state-space models. An enhanced algorithm that is capable of accurately approximating the Bayesian optimal smoother is proposed. It utilizes the exact expression for the quotient of two Gaussian densities to help solve the backward-time recursive equations of Bayesian smoothing, and computes the joint posterior of the state vector and model index. The proposed algorithm only involves the approximation of each model-matched state posterior, which is a Gaussian mixture, with a single Gaussian density for maintaining computational tractability in retrodiction. The validity of the newly developed smoother is verified using a simulated maneuvering target tracking task.
{"title":"Enhanced Fixed-Interval Smoothing for Markovian Switching Systems","authors":"Xi Li, Yi Liu, Le Yang, L. Mihaylova, Bing Deng","doi":"10.23919/fusion49465.2021.9626863","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626863","url":null,"abstract":"This paper considers the problem of fixed-interval smoothing for Markovian switching systems with multiple linear state-space models. An enhanced algorithm that is capable of accurately approximating the Bayesian optimal smoother is proposed. It utilizes the exact expression for the quotient of two Gaussian densities to help solve the backward-time recursive equations of Bayesian smoothing, and computes the joint posterior of the state vector and model index. The proposed algorithm only involves the approximation of each model-matched state posterior, which is a Gaussian mixture, with a single Gaussian density for maintaining computational tractability in retrodiction. The validity of the newly developed smoother is verified using a simulated maneuvering target tracking task.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"50 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125714027","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}