Pub Date : 2021-11-01DOI: 10.23919/fusion49465.2021.9626915
J. Z. Hare, L.M. Kaplan, V. Veeravalli
We study the problem of Quickest Change Detection (QCD) where the parameters of both the pre- and post-change distributions are completely unknown or known within a second-order distribution generated from training data. We propose the use of the Uncertain Likelihood Ratio (ULR) test statistic, which is designed from a Bayesian perspective in contrast with the traditional frequentist approach, i.e., the Generalized Likelihood Ratio (GLR) test. The ULR test utilizes a ratio of posterior predictive distributions, which incorporates parameter uncertainty into the likelihood estimates when there is a lack of or limited availability of training samples. Through an empirical study, we show that the proposed test outperforms the GLR test, while achieving similar results as the classical CUSUM algorithm as the number of training samples goes to infinity.
{"title":"Toward Uncertainty Aware Quickest Change Detection","authors":"J. Z. Hare, L.M. Kaplan, V. Veeravalli","doi":"10.23919/fusion49465.2021.9626915","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626915","url":null,"abstract":"We study the problem of Quickest Change Detection (QCD) where the parameters of both the pre- and post-change distributions are completely unknown or known within a second-order distribution generated from training data. We propose the use of the Uncertain Likelihood Ratio (ULR) test statistic, which is designed from a Bayesian perspective in contrast with the traditional frequentist approach, i.e., the Generalized Likelihood Ratio (GLR) test. The ULR test utilizes a ratio of posterior predictive distributions, which incorporates parameter uncertainty into the likelihood estimates when there is a lack of or limited availability of training samples. Through an empirical study, we show that the proposed test outperforms the GLR test, while achieving similar results as the classical CUSUM algorithm as the number of training samples goes to infinity.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"1 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":"128763956","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/fusion49465.2021.9626925
Nesrine Harbaoui, Khoder Makkawi, Nourdine Ait Tmazirte, Maan El Badaoui El Najjar
For a localization function, meeting together safety, accuracy and availability is a challenging task. Targeting one of these Key Performance Indicators (KPIs) remains feasible but when one or more other requirements are expected at the same time, the objectives become antagonistic. To achieve accuracy, a multi-sensor data fusion is recommended. However, it remains insufficient when it comes to safety critical applications as autonomous vehicle. Indeed, a diagnostic layer has to be considered to treat the presence of faults in dynamic environment, which can affect the sensors measurements. The detection algorithm must ensure high fault sensitivity while keeping false alarm rate as low as possible and taking into account both the change of navigation context and the change of targeted KPIs. This paper proposes a GNSS (Global Navigation Satellite System) and INS (Inertial Navigation system) data fusion approach based on an unscented information filter for state estimation boosted by an adaptive diagnostic layer consisting of a Fault Detection and Isolation (FDI) method based on a powerful parametric information divergence: the α-Rényi divergence. The concept of diagnosis adaptability is developed by applying a sigmoïd strategy in order to increase the sensitivity of the selected residual to detect maximum of faults according to the crossed environment. The suitable selection, at each instant, of α, is ensured through the implementation of a generalized logistic function according to the current constraint of the navigation context. Following the detection step, a decision-cost optimized threshold is reevaluated at each instant. Applied to field data, the first experiments show promising results of the developed framework compared to a diagnostic layer based on the well-known Kullback-Leibler divergence.
{"title":"An α-Rényi Divergence Sigmoïd Parametrization For a Multi-Objectives and Context-Adaptive Fault Tolerant Localization","authors":"Nesrine Harbaoui, Khoder Makkawi, Nourdine Ait Tmazirte, Maan El Badaoui El Najjar","doi":"10.23919/fusion49465.2021.9626925","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626925","url":null,"abstract":"For a localization function, meeting together safety, accuracy and availability is a challenging task. Targeting one of these Key Performance Indicators (KPIs) remains feasible but when one or more other requirements are expected at the same time, the objectives become antagonistic. To achieve accuracy, a multi-sensor data fusion is recommended. However, it remains insufficient when it comes to safety critical applications as autonomous vehicle. Indeed, a diagnostic layer has to be considered to treat the presence of faults in dynamic environment, which can affect the sensors measurements. The detection algorithm must ensure high fault sensitivity while keeping false alarm rate as low as possible and taking into account both the change of navigation context and the change of targeted KPIs. This paper proposes a GNSS (Global Navigation Satellite System) and INS (Inertial Navigation system) data fusion approach based on an unscented information filter for state estimation boosted by an adaptive diagnostic layer consisting of a Fault Detection and Isolation (FDI) method based on a powerful parametric information divergence: the α-Rényi divergence. The concept of diagnosis adaptability is developed by applying a sigmoïd strategy in order to increase the sensitivity of the selected residual to detect maximum of faults according to the crossed environment. The suitable selection, at each instant, of α, is ensured through the implementation of a generalized logistic function according to the current constraint of the navigation context. Following the detection step, a decision-cost optimized threshold is reevaluated at each instant. Applied to field data, the first experiments show promising results of the developed framework compared to a diagnostic layer based on the well-known Kullback-Leibler divergence.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"148 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":"116338190","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/fusion49465.2021.9626898
Keith A. LeGrand, Pingping Zhu, S. Ferrari
Through automatic control, intelligent sensors can be manipulated to obtain the most informative measurements about objects in their environment. In object tracking applications, sensor actions are chosen based on the predicted improvement in estimation accuracy, or information gain. Although random finite set theory provides a formalism for measuring information gain for multi-object tracking problems, predicting the information gain remains computationally challenging. This paper presents a new tractable approximation of the random finite set expected information gain applicable to multi-object search and tracking. The approximation presented in this paper accounts for noisy measurements, missed detections, false alarms, and object appearance/disappearance. The effectiveness of the approach is demonstrated through a ground vehicle tracking problem using real video data from a remote optical sensor.
{"title":"A Random Finite Set Sensor Control Approach for Vision-based Multi-object Search-While-Tracking","authors":"Keith A. LeGrand, Pingping Zhu, S. Ferrari","doi":"10.23919/fusion49465.2021.9626898","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626898","url":null,"abstract":"Through automatic control, intelligent sensors can be manipulated to obtain the most informative measurements about objects in their environment. In object tracking applications, sensor actions are chosen based on the predicted improvement in estimation accuracy, or information gain. Although random finite set theory provides a formalism for measuring information gain for multi-object tracking problems, predicting the information gain remains computationally challenging. This paper presents a new tractable approximation of the random finite set expected information gain applicable to multi-object search and tracking. The approximation presented in this paper accounts for noisy measurements, missed detections, false alarms, and object appearance/disappearance. The effectiveness of the approach is demonstrated through a ground vehicle tracking problem using real video data from a remote optical sensor.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"EM-27 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":"126525193","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/fusion49465.2021.9626917
Yuanna Liu, Jie Geng, Xinyang Deng, Wen Jiang
Cross-lingual entity alignment refers to linking entities in different language knowledge graphs if they are of identical meaning. Recent works focus on learning structure information of knowledge graphs and calculate the distance of entity embeddings for entity alignment. However, the GCN-based methods may bring noise from neighbors due to the heterogeneity of knowledge graphs. Besides, relations, as inherent attribute of knowledge graph, should be merged into the structure learning. In this paper, a relation-aware neighborhood aggregation model RANA is proposed to solve cross-lingual entity alignment task. The specific relation semantics are modeled to modify the aggregation weights of neighbors. CSLS and knowledge graph completion are introduced to enhance the alignment metric and structural information respectively. Experiments on real-world datasets demonstrate that RANA significantly outperforms other baselines in alignment accuracy and robustness.
{"title":"Relation-Aware Neighborhood Aggregation for Cross-lingual Entity Alignment","authors":"Yuanna Liu, Jie Geng, Xinyang Deng, Wen Jiang","doi":"10.23919/fusion49465.2021.9626917","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626917","url":null,"abstract":"Cross-lingual entity alignment refers to linking entities in different language knowledge graphs if they are of identical meaning. Recent works focus on learning structure information of knowledge graphs and calculate the distance of entity embeddings for entity alignment. However, the GCN-based methods may bring noise from neighbors due to the heterogeneity of knowledge graphs. Besides, relations, as inherent attribute of knowledge graph, should be merged into the structure learning. In this paper, a relation-aware neighborhood aggregation model RANA is proposed to solve cross-lingual entity alignment task. The specific relation semantics are modeled to modify the aggregation weights of neighbors. CSLS and knowledge graph completion are introduced to enhance the alignment metric and structural information respectively. Experiments on real-world datasets demonstrate that RANA significantly outperforms other baselines in alignment accuracy and robustness.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"122 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":"127072653","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/fusion49465.2021.9626897
Thies de Boer, M. Schöpe, H. Driessen
The radar resource management problem in a multi-target tracking scenario is considered. Partially observable Markov decision processes (POMDPs) are used to describe each tracking task. Model predictive control is applied to solve the POMDPs in a non-myopic way. As a result, the computational complexity compared to stochastic optimization methods such as policy rollout is dramatically reduced while the resource allocation results maintain similar. This is shown through simulations of dynamic multi-target tracking scenarios in which the cost and computational complexity of different approaches are compared.
{"title":"Radar Resource Management for Multi-Target Tracking Using Model Predictive Control","authors":"Thies de Boer, M. Schöpe, H. Driessen","doi":"10.23919/fusion49465.2021.9626897","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626897","url":null,"abstract":"The radar resource management problem in a multi-target tracking scenario is considered. Partially observable Markov decision processes (POMDPs) are used to describe each tracking task. Model predictive control is applied to solve the POMDPs in a non-myopic way. As a result, the computational complexity compared to stochastic optimization methods such as policy rollout is dramatically reduced while the resource allocation results maintain similar. This is shown through simulations of dynamic multi-target tracking scenarios in which the cost and computational complexity of different approaches are compared.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"75 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":"133656590","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/fusion49465.2021.9627067
Michaela Lobato, W. Norris, R. Nagi, A. Soylemezoglu, Dustin Nottage
Soil moisture content is a key component in terrain characterization for site selection and trafficability assessment. It is laborious and time-consuming to determine soil moisture content using traditional in situ soil moisture sensing methods and may be infeasible for large or dangerous sites. By employing remote sensing techniques, soil moisture content can be determined in a safe and efficient manner. In this work, the results of Keller et al. [1] are expanded upon by reducing the dimensionality of a hyperspectral dataset, resulting in an increase in soil moisture content prediction accuracy. Ten models were developed to predict soil moisture – two machine learning models, support vector machine (SVM) and extremely randomized trees (ET), were trained on 5 input variables. The results indicated that soil moisture content could be predicted with greater accuracy by reducing the dimensionality of a hyperspectral dataset to resemble a standard multispectral dataset. The validity of this method is confirmed by creating a multispectral dataset and concatenating it to the reduced dimensionality (RD) set for an accuracy increase. The ET model’s estimates of soil moisture content outperformed the baseline hyperspectral dataset: obtaining an increase of 1.3% and 5.4% in R-squared values (with a corresponding decrease of .13 and .22 in mean absolute error MAE) when trained on RD and concatenated multispectral (CM) datasets, respectively.
{"title":"Machine Learning for Soil Moisture Prediction Using Hyperspectral and Multispectral Data","authors":"Michaela Lobato, W. Norris, R. Nagi, A. Soylemezoglu, Dustin Nottage","doi":"10.23919/fusion49465.2021.9627067","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9627067","url":null,"abstract":"Soil moisture content is a key component in terrain characterization for site selection and trafficability assessment. It is laborious and time-consuming to determine soil moisture content using traditional in situ soil moisture sensing methods and may be infeasible for large or dangerous sites. By employing remote sensing techniques, soil moisture content can be determined in a safe and efficient manner. In this work, the results of Keller et al. [1] are expanded upon by reducing the dimensionality of a hyperspectral dataset, resulting in an increase in soil moisture content prediction accuracy. Ten models were developed to predict soil moisture – two machine learning models, support vector machine (SVM) and extremely randomized trees (ET), were trained on 5 input variables. The results indicated that soil moisture content could be predicted with greater accuracy by reducing the dimensionality of a hyperspectral dataset to resemble a standard multispectral dataset. The validity of this method is confirmed by creating a multispectral dataset and concatenating it to the reduced dimensionality (RD) set for an accuracy increase. The ET model’s estimates of soil moisture content outperformed the baseline hyperspectral dataset: obtaining an increase of 1.3% and 5.4% in R-squared values (with a corresponding decrease of .13 and .22 in mean absolute error MAE) when trained on RD and concatenated multispectral (CM) datasets, respectively.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"28 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":"133850542","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/fusion49465.2021.9626847
E. Blasch, J. Pieter de Villiers, Gregor Pavin, A. Jousselme, P. Costa, Kathryn B. Laskey, J. Ziegler
The EXCITE (EXplainability Capability Information Testing Evaluation) approach assesses information fusion interpretability, explainability, and accountability for uncertainty analysis. Amongst many data and information fusion techniques is the need to understand the information fusion system capability for the intended application. While many approaches for data fusion support uncertainty reduction from measured data, there are other contributing factors such as data source credibility, knowledge completeness, multiresolution, and problem alignment. To facilitate the alignment of the data fusion approach to the user’s intended action, there is a need towards a representation of the uncertainty. The paper highlights the approach to leverage recent research efforts in interpretability as methods of data handing in the Uncertainty Representation and Reasoning Evaluation Framework (URREF) while also proposing explainability and accountability as a representation criterion. Accountability is closely associated with the selected decision and the outcome which has these four attributes: amount of data towards the result, distance score of decision selection, accuracy/credibility/timeliness of results, and risk analysis. The risk analysis includes: verifiability, observability, liability, transparency, attributability, and remediabilty. Results are demonstrated on notional example.
{"title":"Use of the URREF towards Information Fusion Accountability Evaluation","authors":"E. Blasch, J. Pieter de Villiers, Gregor Pavin, A. Jousselme, P. Costa, Kathryn B. Laskey, J. Ziegler","doi":"10.23919/fusion49465.2021.9626847","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626847","url":null,"abstract":"The EXCITE (EXplainability Capability Information Testing Evaluation) approach assesses information fusion interpretability, explainability, and accountability for uncertainty analysis. Amongst many data and information fusion techniques is the need to understand the information fusion system capability for the intended application. While many approaches for data fusion support uncertainty reduction from measured data, there are other contributing factors such as data source credibility, knowledge completeness, multiresolution, and problem alignment. To facilitate the alignment of the data fusion approach to the user’s intended action, there is a need towards a representation of the uncertainty. The paper highlights the approach to leverage recent research efforts in interpretability as methods of data handing in the Uncertainty Representation and Reasoning Evaluation Framework (URREF) while also proposing explainability and accountability as a representation criterion. Accountability is closely associated with the selected decision and the outcome which has these four attributes: amount of data towards the result, distance score of decision selection, accuracy/credibility/timeliness of results, and risk analysis. The risk analysis includes: verifiability, observability, liability, transparency, attributability, and remediabilty. Results are demonstrated on notional example.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"106 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":"115546048","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/fusion49465.2021.9626976
Li Li, Qinchen Wu, Bin Yan, Shaoming Wei, Jun Wang
Multi-target tracking is an extremely challenging task when targets move in the formation of groups and interact with each other. Group target tracking has to deal with this problem in contrast to independently moving targets as assumed in most multi-target tracking algorithms. A feasible approach for group target tracking is to estimate the group structure and modify the motion model in the prediction step of multi-target tracker according to the group structure. In this paper, we propose an ad hoc labeled multi-Bernoulli (LMB) filter for tracking group target with interaction, which use stochastic differential equation to model the joint motion of group targets and estimate group structure by using graph theory. Simulation results show that the proposed algorithm can estimate the target state more accurately than the traditional method without group motion modification.
{"title":"Labeled Multi-Bernoulli Filter based Group Target Tracking Using SDE and Graph Theory","authors":"Li Li, Qinchen Wu, Bin Yan, Shaoming Wei, Jun Wang","doi":"10.23919/fusion49465.2021.9626976","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626976","url":null,"abstract":"Multi-target tracking is an extremely challenging task when targets move in the formation of groups and interact with each other. Group target tracking has to deal with this problem in contrast to independently moving targets as assumed in most multi-target tracking algorithms. A feasible approach for group target tracking is to estimate the group structure and modify the motion model in the prediction step of multi-target tracker according to the group structure. In this paper, we propose an ad hoc labeled multi-Bernoulli (LMB) filter for tracking group target with interaction, which use stochastic differential equation to model the joint motion of group targets and estimate group structure by using graph theory. Simulation results show that the proposed algorithm can estimate the target state more accurately than the traditional method without group motion modification.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"390 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":"115914237","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/fusion49465.2021.9626967
Mengjiao Tang, Yao Rong, Chen Chen
A popular method for fusing a set of covariance matrix estimates (with unavailable correlation) is to solve their geometrical mean or median, which is defined by a Riemannian geometry of Hermitian positive-definite (HPD) matrices. The most well-known such geometry is identical to the Fisher information geometry of multivariate Gaussian distributions with a fixed mean. This paper identifies the space of HPD matrices with the manifold of centered (i.e., zero-mean) complex elliptically symmetric (CES) distributions. First, the Fisher information matrix for the CES distributions defines a different Riemannian metric on HPD matrices, and the induced Riemannian geometry is studied. Then, the Riemannian Lp mean of some HPD matrices is calculated to produce a final estimation for the scatter matrix (proportional to the covariance matrix) of a CES distribution. While the corresponding objective function is proven to be gconvex, a Riemannian gradient descent algorithm is given to compute the solution. Finally, numerical examples are provided to illustrate the derived geometrical structure and its application to target detection.
{"title":"Riemannian Lp Center of Mass for Scatter Matrix Estimation in Complex Elliptically Symmetric Distributions","authors":"Mengjiao Tang, Yao Rong, Chen Chen","doi":"10.23919/fusion49465.2021.9626967","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626967","url":null,"abstract":"A popular method for fusing a set of covariance matrix estimates (with unavailable correlation) is to solve their geometrical mean or median, which is defined by a Riemannian geometry of Hermitian positive-definite (HPD) matrices. The most well-known such geometry is identical to the Fisher information geometry of multivariate Gaussian distributions with a fixed mean. This paper identifies the space of HPD matrices with the manifold of centered (i.e., zero-mean) complex elliptically symmetric (CES) distributions. First, the Fisher information matrix for the CES distributions defines a different Riemannian metric on HPD matrices, and the induced Riemannian geometry is studied. Then, the Riemannian Lp mean of some HPD matrices is calculated to produce a final estimation for the scatter matrix (proportional to the covariance matrix) of a CES distribution. While the corresponding objective function is proven to be gconvex, a Riemannian gradient descent algorithm is given to compute the solution. Finally, numerical examples are provided to illustrate the derived geometrical structure and its application to target detection.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"122 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":"124818702","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/fusion49465.2021.9626932
Yuxing Yang, Yang Xian, Zeyu Fu, S. M. Naqvi
Video anomaly detection aims to recognise and analyse the video sequences to classify the normal and abnormal frames. This technology can efficiently reduce the human labour to discover the anomalies in surveillance systems and is widely applied in financial, public security and transport sectors. However, video anomaly detection performance is often degraded by the dataset quality, especially for small objects in video sequences. Besides, the computational cost of the classification model would be required as low as possible. In this paper, we proposed information fusion with a joint model which contains motion estimation, object detection and adversarial learning to detect anomalies in two video datasets: UCSD PED1 and PED2. Experimental results confirm the proposed method outperforms the state-of-the-art methods with the additional advantages in reduced computation cost.
{"title":"Video Anomaly Detection for Surveillance Based on Effective Frame Area","authors":"Yuxing Yang, Yang Xian, Zeyu Fu, S. M. Naqvi","doi":"10.23919/fusion49465.2021.9626932","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626932","url":null,"abstract":"Video anomaly detection aims to recognise and analyse the video sequences to classify the normal and abnormal frames. This technology can efficiently reduce the human labour to discover the anomalies in surveillance systems and is widely applied in financial, public security and transport sectors. However, video anomaly detection performance is often degraded by the dataset quality, especially for small objects in video sequences. Besides, the computational cost of the classification model would be required as low as possible. In this paper, we proposed information fusion with a joint model which contains motion estimation, object detection and adversarial learning to detect anomalies in two video datasets: UCSD PED1 and PED2. Experimental results confirm the proposed method outperforms the state-of-the-art methods with the additional advantages in reduced computation cost.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"4 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":"122551409","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}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：