Final Report Summary - TRAX (Training network on tRAcking in compleX sensor systems)
Introduction
Object tracking using sensor data is pervasive in our society with wide spread applications ranging from healthcare, movies, gaming, sports, retail, search and rescue, monitoring and surveillance, and safety and security. Requirements for all these tracking applications are becoming increasingly more demanding. The number of sensors is increasing as is their quality of resolution and the associated amount of data. More demands are placed on the type of objects, their dynamics and appearance in the sensor data to be tracked, calling for much more detailed models to be developed. The number of objects on the scene, as well as the size of the scene is expanding. Without innovative solutions, current tracking solutions will quickly become useless in future complex applications. TRAX is concerned with development of novel techniques for Tracking in Complex Sensor Systems. A total of 15 fellows, 12 ESRs and 3 ERs, have been trained in this extremely challenging field.
Consortium
The TRAX consortium is comprised of a smart mixture of universities, research institutes and small and large companies. The consortium includes the University of Sheffield and Rinicom from UK, Linköping University and Ericsson from Sweden, Fraunhofer FKIE from Germany, and the University of Twente, Xsens, and Thales from The Netherlands. Senionlab (SE) and the University of Bonn (GE) are included as associate partners.
Summary of objectives
The main goal of TRAX is to investigate and design innovative techniques for tracking in complex sensor systems, exploiting the expanding future computational and memory storage capabilities. Advanced dynamic and measurement models for those complex systems have been developed, as well as efficient algorithms exploiting among others model structure, sparseness, particle filtering and parallel and cloud computing. The main training goal of TRAX was to establish a novel research training program on Tracking in Complex Sensor Systems, covering interdisciplinary and intersectoral aspects in this newly emerging supra-disciplinary field.
Description of work performed
After an initial recruitment period, a kick-off was organized to let the research fellows meet and share their initial ideas for their individual research projects. Personal career development plans were compiled and finalized. Four network meetings have been organized as Summer Schools in Linköping (SE, 2014), Sheffield (UK, 2015), Bonn (GE, 2016), and Enschede (Twente, NL, 2017). During the Summers Schools fellows received joint trainings and discussed jointly their individual research results and further plans. A number of fellows have done a secondment at research facilities of other partners in the consortium to perform joint research and profit from their individual strengths.
Main results achieved
During the course of TRAX, already four fellows have obtained their PhD degree, and two have successfully defended their Licentiate Thesis, six fellows are still enrolled as PhD students at various universities, and one fellow is currently working towards his Licentiate Thesis. The fellows have received their individual and joint training, both via on-the-job training, a variety of courses, and the Summer Schools. To be able to deal with huge amounts of data, complex object and sensor models, as well as large amounts of objects, several new models and estimation techniques have been proposed. Amongst the proposed models are “Dynamic Topic Modelling using both parametric as well as non-parametric models”, “the Chinese Restaurant Process”, and “Gaussian Processes”. Amongst the efficient estimation techniques are “Log-Homotopy particle flows”, “Hybrid Monte Carlo (HMC)”, “Box particle filtering”, “Convolutional Particle Filtering (CPF)”, “sub-sampling”, and “Expectation Propagation (EP)”. These models and techniques have been researched for realizing applications involving joint estimation and clustering/classification, and joint state vector and parameter estimation as can be found in situation assessment in assisted driving, tracking crowds of objects, and extended objects, and identifying object behavior and radio propagation in wireless applications.
Impact
It is fair to say that many models and techniques have been developed that make possible improvements in a wide range of applications that beforehand seemed very difficult. In addition to that, we have trained 15 fellows in exploring novel models and techniques to realize important products and applications that are impacting our society already today and will have even more impact in the future.
Website
Further information about TRAX, its application and goals, the universities, research institutes and small and large companies involved, the individual researchers and their projects, as well as the network events, the publications, can be found on the TRAX website https://www.trax.utwente.nl/.
Object tracking using sensor data is pervasive in our society with wide spread applications ranging from healthcare, movies, gaming, sports, retail, search and rescue, monitoring and surveillance, and safety and security. Requirements for all these tracking applications are becoming increasingly more demanding. The number of sensors is increasing as is their quality of resolution and the associated amount of data. More demands are placed on the type of objects, their dynamics and appearance in the sensor data to be tracked, calling for much more detailed models to be developed. The number of objects on the scene, as well as the size of the scene is expanding. Without innovative solutions, current tracking solutions will quickly become useless in future complex applications. TRAX is concerned with development of novel techniques for Tracking in Complex Sensor Systems. A total of 15 fellows, 12 ESRs and 3 ERs, have been trained in this extremely challenging field.
Consortium
The TRAX consortium is comprised of a smart mixture of universities, research institutes and small and large companies. The consortium includes the University of Sheffield and Rinicom from UK, Linköping University and Ericsson from Sweden, Fraunhofer FKIE from Germany, and the University of Twente, Xsens, and Thales from The Netherlands. Senionlab (SE) and the University of Bonn (GE) are included as associate partners.
Summary of objectives
The main goal of TRAX is to investigate and design innovative techniques for tracking in complex sensor systems, exploiting the expanding future computational and memory storage capabilities. Advanced dynamic and measurement models for those complex systems have been developed, as well as efficient algorithms exploiting among others model structure, sparseness, particle filtering and parallel and cloud computing. The main training goal of TRAX was to establish a novel research training program on Tracking in Complex Sensor Systems, covering interdisciplinary and intersectoral aspects in this newly emerging supra-disciplinary field.
Description of work performed
After an initial recruitment period, a kick-off was organized to let the research fellows meet and share their initial ideas for their individual research projects. Personal career development plans were compiled and finalized. Four network meetings have been organized as Summer Schools in Linköping (SE, 2014), Sheffield (UK, 2015), Bonn (GE, 2016), and Enschede (Twente, NL, 2017). During the Summers Schools fellows received joint trainings and discussed jointly their individual research results and further plans. A number of fellows have done a secondment at research facilities of other partners in the consortium to perform joint research and profit from their individual strengths.
Main results achieved
During the course of TRAX, already four fellows have obtained their PhD degree, and two have successfully defended their Licentiate Thesis, six fellows are still enrolled as PhD students at various universities, and one fellow is currently working towards his Licentiate Thesis. The fellows have received their individual and joint training, both via on-the-job training, a variety of courses, and the Summer Schools. To be able to deal with huge amounts of data, complex object and sensor models, as well as large amounts of objects, several new models and estimation techniques have been proposed. Amongst the proposed models are “Dynamic Topic Modelling using both parametric as well as non-parametric models”, “the Chinese Restaurant Process”, and “Gaussian Processes”. Amongst the efficient estimation techniques are “Log-Homotopy particle flows”, “Hybrid Monte Carlo (HMC)”, “Box particle filtering”, “Convolutional Particle Filtering (CPF)”, “sub-sampling”, and “Expectation Propagation (EP)”. These models and techniques have been researched for realizing applications involving joint estimation and clustering/classification, and joint state vector and parameter estimation as can be found in situation assessment in assisted driving, tracking crowds of objects, and extended objects, and identifying object behavior and radio propagation in wireless applications.
Impact
It is fair to say that many models and techniques have been developed that make possible improvements in a wide range of applications that beforehand seemed very difficult. In addition to that, we have trained 15 fellows in exploring novel models and techniques to realize important products and applications that are impacting our society already today and will have even more impact in the future.
Website
Further information about TRAX, its application and goals, the universities, research institutes and small and large companies involved, the individual researchers and their projects, as well as the network events, the publications, can be found on the TRAX website https://www.trax.utwente.nl/.