Stochastic geometry for automatic multiple object detection and tracking in remotely sensed high resolution image sequences
| Date: | Wednesday, 25 November 2015 |
| Time: | 10:30 |
| Place: | Salle Euler Violet, Bâtiment Euler, INRIA Sophia Antipolis Méditerranée, 2004 route des Lucioles, 06902 Sophia Antipolis |
Reviewers:
Prof. Alfred M. Bruckstein, Technion – Israel Institute of Technology, Israel
Prof. Ba-Ngu Vo, Curtin University, Australia
Examiners:
Dr. Jacques Blanc-Talon – Direction Générale de l’Armement, France
Dr. Xavier Descombes – INRIA, France
Dr. Mathias Ortner – Airbus Defense and Space, France
Prof. Michel Schmitt – Institut Mines-Telecom, France
Prof. Daniela Zaharie – West University of Timişoara, Romania
Supervisors:
Prof. Josiane Zerubia – INRIA, France
Abstract:
In this thesis, we combine the methods from probability theory and stochastic geometry to put forward new solutions to the multiple object detection and tracking problem in high resolution remotely sensed image sequences.
First, we develop a spatial marked point process model to detect a pre-defined class of objects (i.e. boats) based on their visual and geometric characteristics. We design a multiple core implementation of the reversible jump MCMC sampler. We show improved results over state of the art methods and competitive detection times.
The very good results obtained motivated us to extend these models to the temporal domain and create a framework based on spatio-temporal marked point process models to jointly detect and track multiple objects in image sequences. We propose the use of simple parametric shapes to describe the appearance of these objects. We build new, dedicated energy based models consisting of several terms that take into account both the image evidence and physical constraints such as object dynamics, track persistence and mutual exclusion. We construct a suitable optimization scheme that allows us to find strong local minima of the proposed highly non-convex energy.
As the simulation of such models comes with a high computational cost, we turn our attention to the recent filter implementations for multiple object tracking, which are known to be less computationally expensive. We propose a hybrid sampler by combining the Kalman filter with the standard Reversible Jump MCMC. High performance computing techniques are also used to increase the computational efficiency of our method. We provide an in-depth analysis of the proposed framework, as well as an extensive comparison with state of the art methods, based on standard multiple object tracking metrics and computational efficiency. This analysis yields a very good detection and tracking performance at the price of an increased complexity of the models. Exhaustive tests have been conducted both on high resolution satellite and microscopy image sequences.
First, we develop a spatial marked point process model to detect a pre-defined class of objects (i.e. boats) based on their visual and geometric characteristics. We design a multiple core implementation of the reversible jump MCMC sampler. We show improved results over state of the art methods and competitive detection times.
The very good results obtained motivated us to extend these models to the temporal domain and create a framework based on spatio-temporal marked point process models to jointly detect and track multiple objects in image sequences. We propose the use of simple parametric shapes to describe the appearance of these objects. We build new, dedicated energy based models consisting of several terms that take into account both the image evidence and physical constraints such as object dynamics, track persistence and mutual exclusion. We construct a suitable optimization scheme that allows us to find strong local minima of the proposed highly non-convex energy.
As the simulation of such models comes with a high computational cost, we turn our attention to the recent filter implementations for multiple object tracking, which are known to be less computationally expensive. We propose a hybrid sampler by combining the Kalman filter with the standard Reversible Jump MCMC. High performance computing techniques are also used to increase the computational efficiency of our method. We provide an in-depth analysis of the proposed framework, as well as an extensive comparison with state of the art methods, based on standard multiple object tracking metrics and computational efficiency. This analysis yields a very good detection and tracking performance at the price of an increased complexity of the models. Exhaustive tests have been conducted both on high resolution satellite and microscopy image sequences.
Keywords:
Multiple object tracking, object detection, marked point process, Kalman filter, satellite image sequences, microscopy data sequences, high resolution.
