Pub Date : 2021-11-01DOI: 10.23919/fusion49465.2021.9626943
J. Buyer, Dominic Waldenmayer, R. Zöllner, Johann Marius Zöllner
The paper presents a prediction framework applied to vehicle speed prediction in multi-lane traffic. In general, the framework combines components of (heuristic) model knowledge, probabilistic estimation and machine learning techniques in order to benefit from the different advantages of the respective methods. The idea of the approach is merging several models suitable for simple environments (called basis models) to a bigger model suitable for complex environments (called overall model). Since the number of basis models is limited due to faster execution time, system boundaries are determined, which implicates a selection of the relevant model inputs. For improved prediction performance, the values of the model parameters are estimated online via recursive Bayesian estimation. Moreover, data-driven models are integrated for adaptive weighting of the basis models in order to represent time-varying behavior over the prediction horizon. In the application of the framework to vehicle speed prediction, single-lane car-following models are used as basis models. The determination of the system boundaries is based on a decision tree. Model parameter estimation is realized with a particle filter implementation and the data-driven models for weighting the basis models are realized as support vector machine (SVM) regression models. Experimental results of the suggested vehicle speed prediction framework show an improved performance of the approach compared to a related baseline approach.
{"title":"Data-Driven Merging of Car-Following Models for Interaction-Aware Vehicle Speed Prediction","authors":"J. Buyer, Dominic Waldenmayer, R. Zöllner, Johann Marius Zöllner","doi":"10.23919/fusion49465.2021.9626943","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626943","url":null,"abstract":"The paper presents a prediction framework applied to vehicle speed prediction in multi-lane traffic. In general, the framework combines components of (heuristic) model knowledge, probabilistic estimation and machine learning techniques in order to benefit from the different advantages of the respective methods. The idea of the approach is merging several models suitable for simple environments (called basis models) to a bigger model suitable for complex environments (called overall model). Since the number of basis models is limited due to faster execution time, system boundaries are determined, which implicates a selection of the relevant model inputs. For improved prediction performance, the values of the model parameters are estimated online via recursive Bayesian estimation. Moreover, data-driven models are integrated for adaptive weighting of the basis models in order to represent time-varying behavior over the prediction horizon. In the application of the framework to vehicle speed prediction, single-lane car-following models are used as basis models. The determination of the system boundaries is based on a decision tree. Model parameter estimation is realized with a particle filter implementation and the data-driven models for weighting the basis models are realized as support vector machine (SVM) regression models. Experimental results of the suggested vehicle speed prediction framework show an improved performance of the approach compared to a related baseline approach.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"59 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":"127368581","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/fusion49465.2021.9626992
Qichao Tang, Z. Duan
In multi-sensor distributed estimation fusion, local estimation errors are generally correlated among local estimates. Usually, correlation is known to exist but unavailable or unclear to be how large it is, and needed to be considered. For this situation, a sensible way is to set up an optimality criterion and optimize it over all possible such correlations. Based on the framework of minimizing the statistical distance sum between the fused density and local posterior densities, a new method is proposed by utilizing Bhattacharyya distance, which is commonly used in measuring the closeness or similarity between two densities. First, the objective function is introduced. Then, we investigate the convexity form of the objective function, and separate the solving procedure into two steps during settling the original optimization problem, which benefit us to acquire the solution of that problem. At last, the acquired solution (fused estimate) is given in an implicit form, however, it can be obtained through iterative algorithm. And, it is pessimistic definite in mean square error (MSE). Numerical examples illustrate this and show the effectiveness of the proposed distributed method by comparing with several other fusion methods under the same framework but using other kinds of statistical distance.
{"title":"Multi-sensor Distributed Estimation Fusion Based on Minimizing the Bhattacharyya Distance Sum","authors":"Qichao Tang, Z. Duan","doi":"10.23919/fusion49465.2021.9626992","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626992","url":null,"abstract":"In multi-sensor distributed estimation fusion, local estimation errors are generally correlated among local estimates. Usually, correlation is known to exist but unavailable or unclear to be how large it is, and needed to be considered. For this situation, a sensible way is to set up an optimality criterion and optimize it over all possible such correlations. Based on the framework of minimizing the statistical distance sum between the fused density and local posterior densities, a new method is proposed by utilizing Bhattacharyya distance, which is commonly used in measuring the closeness or similarity between two densities. First, the objective function is introduced. Then, we investigate the convexity form of the objective function, and separate the solving procedure into two steps during settling the original optimization problem, which benefit us to acquire the solution of that problem. At last, the acquired solution (fused estimate) is given in an implicit form, however, it can be obtained through iterative algorithm. And, it is pessimistic definite in mean square error (MSE). Numerical examples illustrate this and show the effectiveness of the proposed distributed method by comparing with several other fusion methods under the same framework but using other kinds of statistical distance.","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":"128922823","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/fusion49465.2021.9627020
Ashton Harvey, Kathryn B. Laskey, Kuo-Chu Chang
Space is becoming a more crowded and contested domain, but the techniques used to task the sensors monitoring this environment have not significantly changed since the implementation of James Miller’s marginal analysis technique used in the Special Perturbations (SP) Tasker in 2007. Centralized tasker / scheduler approaches have used a Markov Decision Process (MDP) formulation, but myopic solutions fail to account for future states and non-myopic solutions tend to be computationally infeasible at scale. Linares and Furfaro proposed solving an MDP formulation of the Sensor Allocation Problem (SAP) using Deep Reinforcement Learning (DRL). DRL has been instrumental in solving many high-dimensional control problems previously considered too complex to solve at an expert level, including Go, Atari 2600, Dota 2, Starcraft 2 and autonomous driving. Linares and Furfaro showed DRL could converge on effective policies for sets of up to 300 objects in the same orbital plane. Jones expanded on that work to a full three-dimensional case with objects in diverse orbits. DRL methods can require significant training time to learn from an a priori state. This paper builds on past work by applying imitation learning to bootstrap DRL methods with existing heuristic solutions. We show that a Demonstration Guided DRL (DG-DRL) approach can effectively replicate a near-optimal tasker’s performance using trajectories from a sub-optimal heuristic. Further, we show that our approach avoids the poor initial performance typical of online DRL approaches. Code is available as an open source library at: https://github.com/AshHarvey/ssa-gym
{"title":"Bridging Heuristic and Deep Learning Approaches to Sensor Tasking","authors":"Ashton Harvey, Kathryn B. Laskey, Kuo-Chu Chang","doi":"10.23919/fusion49465.2021.9627020","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9627020","url":null,"abstract":"Space is becoming a more crowded and contested domain, but the techniques used to task the sensors monitoring this environment have not significantly changed since the implementation of James Miller’s marginal analysis technique used in the Special Perturbations (SP) Tasker in 2007. Centralized tasker / scheduler approaches have used a Markov Decision Process (MDP) formulation, but myopic solutions fail to account for future states and non-myopic solutions tend to be computationally infeasible at scale. Linares and Furfaro proposed solving an MDP formulation of the Sensor Allocation Problem (SAP) using Deep Reinforcement Learning (DRL). DRL has been instrumental in solving many high-dimensional control problems previously considered too complex to solve at an expert level, including Go, Atari 2600, Dota 2, Starcraft 2 and autonomous driving. Linares and Furfaro showed DRL could converge on effective policies for sets of up to 300 objects in the same orbital plane. Jones expanded on that work to a full three-dimensional case with objects in diverse orbits. DRL methods can require significant training time to learn from an a priori state. This paper builds on past work by applying imitation learning to bootstrap DRL methods with existing heuristic solutions. We show that a Demonstration Guided DRL (DG-DRL) approach can effectively replicate a near-optimal tasker’s performance using trajectories from a sub-optimal heuristic. Further, we show that our approach avoids the poor initial performance typical of online DRL approaches. Code is available as an open source library at: https://github.com/AshHarvey/ssa-gym","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"35 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":"124465413","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/fusion49465.2021.9626882
M. Ulmke
The concept of indistinguishability in multi-target tracking leads to correlations in their statistical description even without explicit interactions between the objects. These correlations can be described in terms of a wave function – the square-root of the multi-target probability density function (pdf) – which is necessarily either symmetric or anti-symmetric under the exchange of two target indices. [1] This symmetry dichotomy, well-known in quantum many particle physics as bosonic and fermionic behavior, leads to specific properties of the multi-target pdf. While anti-symmetry results in a repulsive behavior in terms of the single object states, symmetry leads to clustering of single objects into the same state. This different behavior can be exploited to describe macroscopic objects which either tend to avoid each other or to form groups.In this paper, we develop an approach for tracking multiple non-interacting indistinguishable targets in the presence of false alarms. The goal is to avoid the treatment of the high-dimensional multi-target pdf by approximating it in terms of the square of so-called Slater determinants and permanents build from single target pdfs. From the intensity (first order statistical moment) of the multi-target pdf, we derive approximations for single target pdfs which show the specific fermionic and bosonic behavior. These "corrected" single target pdfs can serve as input into standard data association and filtering algorithms. Exemplary implementations in a JPDAF framework demonstrate the mitigation of track coalescence.
{"title":"Single-target density for tracking indistinguishable objects","authors":"M. Ulmke","doi":"10.23919/fusion49465.2021.9626882","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626882","url":null,"abstract":"The concept of indistinguishability in multi-target tracking leads to correlations in their statistical description even without explicit interactions between the objects. These correlations can be described in terms of a wave function – the square-root of the multi-target probability density function (pdf) – which is necessarily either symmetric or anti-symmetric under the exchange of two target indices. [1] This symmetry dichotomy, well-known in quantum many particle physics as bosonic and fermionic behavior, leads to specific properties of the multi-target pdf. While anti-symmetry results in a repulsive behavior in terms of the single object states, symmetry leads to clustering of single objects into the same state. This different behavior can be exploited to describe macroscopic objects which either tend to avoid each other or to form groups.In this paper, we develop an approach for tracking multiple non-interacting indistinguishable targets in the presence of false alarms. The goal is to avoid the treatment of the high-dimensional multi-target pdf by approximating it in terms of the square of so-called Slater determinants and permanents build from single target pdfs. From the intensity (first order statistical moment) of the multi-target pdf, we derive approximations for single target pdfs which show the specific fermionic and bosonic behavior. These \"corrected\" single target pdfs can serve as input into standard data association and filtering algorithms. Exemplary implementations in a JPDAF framework demonstrate the mitigation of track coalescence.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"16 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":"114180759","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/fusion49465.2021.9627033
Joshua Gehlen, F. Govaers, W. Koch
Closely spaced multi target tracking remains a challenging problem in state estimation and data fusion. A recent formulation of the problem using antisymmetric square roots of density functions, which may be interpreted as multi target wave functions, has proposed a separation of densities by means of the resulting "Pauli-Notch". In this paper, this formulation is extended for non-Gaussian posterior densities, which are given in discretized and Candecomp-/Parafac decomposed form. Such densities can be predicted by a numerical solution of the Fokker-Planck-Equation. A modified operator for the respective wave function is presented together with the Bayes recursion in order to solve state estimation based on antisymmetric wave functions.
{"title":"On Tracking Closely-Spaced Targets in a PARAFAC-Representation of the Fermionic Wave Function Formulation","authors":"Joshua Gehlen, F. Govaers, W. Koch","doi":"10.23919/fusion49465.2021.9627033","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9627033","url":null,"abstract":"Closely spaced multi target tracking remains a challenging problem in state estimation and data fusion. A recent formulation of the problem using antisymmetric square roots of density functions, which may be interpreted as multi target wave functions, has proposed a separation of densities by means of the resulting \"Pauli-Notch\". In this paper, this formulation is extended for non-Gaussian posterior densities, which are given in discretized and Candecomp-/Parafac decomposed form. Such densities can be predicted by a numerical solution of the Fokker-Planck-Equation. A modified operator for the respective wave function is presented together with the Bayes recursion in order to solve state estimation based on antisymmetric wave functions.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"91 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":"127323211","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/fusion49465.2021.9627013
G. Pavlin, J. D. Villiers, J. Ziegler, A. Jousselme, P. Costa, Kathryn B. Laskey, A. D. Waal, E. Blasch, L. Jansen
Explainability is generally considered an important means to gain trust in complex automated decision support systems. Different types of explainability of processes and models used in a complex information fusion solution based on Artificial Intelligence (AI) are relevant throughout its life-cycle, i.e. during the system development as well as its deployment. However, it is often difficult to understand the real value of explainability in specific cases. To study the impact of explainability on trust, there is a need to emphasize the trust building processes, especially various types of evaluations supporting trust assessment. The paper emphasizes that the value of explainability is as an enabler of certain types of evaluations leading to improved trust in automated solutions. A conceptual model brings together different types of explainability, evaluations, and operational conditions along with human factors influencing the trust in automated systems. The introduced model describes the types of possible evaluations and related explainability at different stages of life cycles of AI-based information fusion solutions. This enables adaptation of life cycles, such that the trust assessment is facilitated. The concepts are illustrated with the help of examples using different modelling and processing techniques.
{"title":"Relations Between Explainability, Evaluation and Trust in AI-Based Information Fusion Systems","authors":"G. Pavlin, J. D. Villiers, J. Ziegler, A. Jousselme, P. Costa, Kathryn B. Laskey, A. D. Waal, E. Blasch, L. Jansen","doi":"10.23919/fusion49465.2021.9627013","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9627013","url":null,"abstract":"Explainability is generally considered an important means to gain trust in complex automated decision support systems. Different types of explainability of processes and models used in a complex information fusion solution based on Artificial Intelligence (AI) are relevant throughout its life-cycle, i.e. during the system development as well as its deployment. However, it is often difficult to understand the real value of explainability in specific cases. To study the impact of explainability on trust, there is a need to emphasize the trust building processes, especially various types of evaluations supporting trust assessment. The paper emphasizes that the value of explainability is as an enabler of certain types of evaluations leading to improved trust in automated solutions. A conceptual model brings together different types of explainability, evaluations, and operational conditions along with human factors influencing the trust in automated systems. The introduced model describes the types of possible evaluations and related explainability at different stages of life cycles of AI-based information fusion solutions. This enables adaptation of life cycles, such that the trust assessment is facilitated. The concepts are illustrated with the help of examples using different modelling and processing techniques.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"21 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":"127257089","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/fusion49465.2021.9627039
Jaya Shradha Fowdur, M. Baum, F. Heymann
Ellipses are favourable when it comes to tracking the shape of targets in a wide range of applications. With enhanced sensor technologies, the need for efficient measurement processing and accurate estimation keeps getting more pronounced. In this paper, we propose an approach called Principal Axes Kalman Filter (PAKF) for tracking an elliptical extended target whose extent parameters are estimated directly from explicit elliptical measurements (lengths of semi-axes and orientation), that have in turn been computed from a high number of (noisy) sensor measurements. The benefits of the approach, both in terms of processing and accuracy, are demonstrated by a comparison with two existing approaches: the random matrix model (RMM) and the Multiplicative Error Model-Extended Kalman Filter* (MEM-EKF*). Moreover, the approach is applied on a real-world standard on-board marine radar dataset and the outcomes are presented and discussed.
{"title":"An Elliptical Principal Axes-based Model for Extended Target Tracking with Marine Radar Data","authors":"Jaya Shradha Fowdur, M. Baum, F. Heymann","doi":"10.23919/fusion49465.2021.9627039","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9627039","url":null,"abstract":"Ellipses are favourable when it comes to tracking the shape of targets in a wide range of applications. With enhanced sensor technologies, the need for efficient measurement processing and accurate estimation keeps getting more pronounced. In this paper, we propose an approach called Principal Axes Kalman Filter (PAKF) for tracking an elliptical extended target whose extent parameters are estimated directly from explicit elliptical measurements (lengths of semi-axes and orientation), that have in turn been computed from a high number of (noisy) sensor measurements. The benefits of the approach, both in terms of processing and accuracy, are demonstrated by a comparison with two existing approaches: the random matrix model (RMM) and the Multiplicative Error Model-Extended Kalman Filter* (MEM-EKF*). Moreover, the approach is applied on a real-world standard on-board marine radar dataset and the outcomes are presented and discussed.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"33 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":"122255089","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/fusion49465.2021.9626918
N. Rao
Machine learning (ML) methods are increasingly being applied to solve complex, data-driven problems in diverse areas, by exploiting the physical laws derived from first principles such as thermal hydraulics and the abstract laws developed recently for data and computing infrastructures. These physical and abstract laws encapsulate, typically in compact algebraic forms, the critical knowledge that complements data-driven ML models. We present a unified perspective of these laws and ML methods using an abstract algebra $(mathcal{A}; oplus , otimes )$, wherein the performance estimation and classification tasks are characterized by the additive ⊕ operations, and the diagnosis, reconstruction, and optimization tasks are characterized by the difference ⊗ operations. This abstraction provides ML codes and their performance characterizations that are transferable across different areas. We describe practical applications of these abstract operations using examples of throughput profile estimation tasks in data transport infrastructures, and power-level and sensor error estimation tasks in nuclear reactor systems.
{"title":"An Algebra of Machine Learners with Applications","authors":"N. Rao","doi":"10.23919/fusion49465.2021.9626918","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626918","url":null,"abstract":"Machine learning (ML) methods are increasingly being applied to solve complex, data-driven problems in diverse areas, by exploiting the physical laws derived from first principles such as thermal hydraulics and the abstract laws developed recently for data and computing infrastructures. These physical and abstract laws encapsulate, typically in compact algebraic forms, the critical knowledge that complements data-driven ML models. We present a unified perspective of these laws and ML methods using an abstract algebra $(mathcal{A}; oplus , otimes )$, wherein the performance estimation and classification tasks are characterized by the additive ⊕ operations, and the diagnosis, reconstruction, and optimization tasks are characterized by the difference ⊗ operations. This abstraction provides ML codes and their performance characterizations that are transferable across different areas. We describe practical applications of these abstract operations using examples of throughput profile estimation tasks in data transport infrastructures, and power-level and sensor error estimation tasks in nuclear reactor systems.","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":"125417475","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/fusion49465.2021.9626933
J. Nordlöf, Gustaf Hendeby, Daniel Axehill
A belief-space planning problem for GNSS-denied areas is studied where the location and number of landmarks available are unknown when performing the planning. To be able to plan an informative path in this situation, an algorithm using virtual landmarks to position the platform during the planning phase is studied. The virtual landmarks are selected to capture the expected information available in different regions of the map, based on the beforehand known landmark density. The main contribution of this work is a better approximation of the obtained information from the virtual landmarks and a theoretical study of the properties of the approximation. Furthermore, the proposed approximation, in itself and its use in a path planner, is investigated with successful results.
{"title":"Improved Virtual Landmark Approximation for Belief-Space Planning","authors":"J. Nordlöf, Gustaf Hendeby, Daniel Axehill","doi":"10.23919/fusion49465.2021.9626933","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626933","url":null,"abstract":"A belief-space planning problem for GNSS-denied areas is studied where the location and number of landmarks available are unknown when performing the planning. To be able to plan an informative path in this situation, an algorithm using virtual landmarks to position the platform during the planning phase is studied. The virtual landmarks are selected to capture the expected information available in different regions of the map, based on the beforehand known landmark density. The main contribution of this work is a better approximation of the obtained information from the virtual landmarks and a theoretical study of the properties of the approximation. Furthermore, the proposed approximation, in itself and its use in a path planner, is investigated with successful results.","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":"129946480","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/fusion49465.2021.9626853
Qing Deng, Wei Tian, Yuyao Huang, Lu Xiong, Xin Bi
Pedestrian detection in low-light environment is an essential part for autonomous driving in all-day and all-weather situations. A current trend is utilizing multispectral information such as RGB and infrared images to detect pedestrians. Despite its efficacy, such an approach suffers from underperformance in dealing with varied object scales due to its limited feature fusion on semantic levels. To address the above problem, we propose a novel multi-layer fusion network called as MLF-FRCNN. In this network, multi-scale feature maps are created from RGB and infrared channels from each backbone block. A feature pyramid network module is further introduced to facilitate predictions on multi-layer feature maps. The experimental results on the KAIST Dataset reveal that our method achieves a runtime performance of 0.14s per frame and an average precision of 91.2% which outperforms state-of-the-art multispectral fusion methods. The effectiveness of our approach in dealing with scaled objects in low-light environment is further proven by ablation studies.
{"title":"Pedestrian Detection by Fusion of RGB and Infrared Images in Low-Light Environment","authors":"Qing Deng, Wei Tian, Yuyao Huang, Lu Xiong, Xin Bi","doi":"10.23919/fusion49465.2021.9626853","DOIUrl":"https://doi.org/10.23919/fusion49465.2021.9626853","url":null,"abstract":"Pedestrian detection in low-light environment is an essential part for autonomous driving in all-day and all-weather situations. A current trend is utilizing multispectral information such as RGB and infrared images to detect pedestrians. Despite its efficacy, such an approach suffers from underperformance in dealing with varied object scales due to its limited feature fusion on semantic levels. To address the above problem, we propose a novel multi-layer fusion network called as MLF-FRCNN. In this network, multi-scale feature maps are created from RGB and infrared channels from each backbone block. A feature pyramid network module is further introduced to facilitate predictions on multi-layer feature maps. The experimental results on the KAIST Dataset reveal that our method achieves a runtime performance of 0.14s per frame and an average precision of 91.2% which outperforms state-of-the-art multispectral fusion methods. The effectiveness of our approach in dealing with scaled objects in low-light environment is further proven by ablation studies.","PeriodicalId":226850,"journal":{"name":"2021 IEEE 24th International Conference on Information Fusion (FUSION)","volume":"36 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":"126998906","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：