ORTNER , Mathias

The context of this thesis is the reconstruction of urban areas from images. It proposes a set of algorithms for extracting simple shapes from Digital Elevation Models (DEM). DEMs describe the altimetry of an urban area by a grid of points, each of which has a height associated to it. The proposed models are based on marked point processes....

Small object tracking in low-resolution remote sensing images presents numerous challenges. Targets are relatively small compared to the field of view, do not present distinct features, and are often lost in cluttered environments. In this paper, we propose a track-by-detection approach to detect and track small moving targets by using a...

The Generalized Labeled Multi-Bernoulli (GLMB) filter attains remarkable results in Multi-Object Tracking (MOT). Nevertheless, the GLMB filter relies on strong assumptions such as prior knowledge of targets' initial state. Pragmatic scenarios such as satellite video object tracking challenge these assumptions as objects appear at random...

This work present an automatic algorithm that extract 3D land register from altimetric data in dense urban areas. Altimetry of a town is a data which is easily available yet difficult to exploit. For instance, we present here results on two kind of measurements : the first one consists in a Digital Elevation Model (DEM) built using a...

Convolutional Neural Networks have shown great results for object detection tasks by learning texture and pattern extraction filters. However, object-level interactions are harder to grasp without increasing the complexity of the architectures. On the other hand Point Process models propose to solve the detection on the configuration of objects...

We present a method combining marked point processes and convolutional neural networks applied to the detection of small objects in optical satellite images. In such images, objects are densely scattered, and visual information is scarce. The point process framework allows factoring in priors to account for object interactions. Classical point...

We present a multi-object tracking (MOT) approach to track small moving targets in satellite images. Our objects of interest span few pixels, do not present a defined texture, and are easily lost in cluttered environments. We propose a patchbased convolutional neural network (CNN) that focuses on specific regions to detect and discriminate...

In this article we present a combination of marked point processes with convolutional neural networks applied to remote sensing. While point processes allow modeling interactions between objects via priors, classical methods rely on contrast measures that become unreliable as objects of interest and context become more diverse. We propose...

This paper presents a model for object detection based on point processes, which allows considering object interactions, while using the features extracted by a convolutional neural network. We also present a contrastive learning method to infer the parameters of the energy model, to then apply our method to the detection of vehicles in optical...

This article presents a method combining marked point processes and convolutional neural networks in order to detect small objects in optical satellite images. In this setting, objects are scattered densely: the energy based formulation of a point process allows us to factor in priors to account for object interactions. Classical marked point...

We first recall Geyer and Møller algorithm that allows to sample point processes using a Markov chain. We also recall Green's framework that allows to build samplers on general state spaces by imposing reversibility of the designed Markov chain.Since in our image processing applications, we are interested by sampling highly spatially correlated...

This work presents a framework for automatic feature extraction from images using stochastic geometry. Features in images are modeled as realizations of a spatial point process of geometrical shapes. This framework allows the incorporation of a priori knowledge on the spatial repartition of features. More specifically, we present a model based...

Cet article présente une méthode combinant processus ponctuels marqués et réseaux de neurones convolutifs pour la détection de petits objets dans des images satellitaires. Dans un tel contexte, la densité des objets est parfois élevée ; la formulation énergétique d'un processus ponctuel nous permet d'intégrer des a priori sur les configurations...

We present a real-time multi-object tracker using an enhanced version of the Gaussian mixture probability hypothesis density (GM-PHD) filter to track detections of a state-of-the-art convolutional neural network (CNN). This approach adapts the GM-PHD filter to a real-world scenario to recover target trajectories in remote sensing videos. Our...

This work presents a framework for automatic feature extraction from images using stochastic geometry. Features in images are modeled as realizations of a spatial point process of geometrical shapes. This framework allows the incorporation of a prior knowledge on the spatial repartition of features. More specifically, we present a model based...

We aim to extract buildings from Digital Elevation Models. To achieve this goal, we define a point process whose points represent buildings. We then define a density for this point process which is split into two parts. When written as an energy this density consists of two fields : the first one is an "internal field" that allows us to model...

In our image processing applications, we use a simulated annealing procedure to find configurations of geometric shapes that fit the best an image. This type of algorithm allows finding one of the global minima of an arbitrary function provided that the cooling schedule is logarithmic with the time. Since this type of cooling schedules is very...

This paper presents a method to combine point processes and convolutional neural networks for detecting small objects in remotely sensed images. In such a context, when objects are small and their density is high, we use priors within a point process simulation. The data term of this point process has been learned with a neural network, thus...

Convolutional neural networks (CNN) have shown great results for object-detection tasks by learning texture and pattern-extraction filters. However, object-level interactions are harder to grasp without increasing the complexity of the architectures. On the other hand, Point Process models propose to solve the detection of the configuration of...

Convolutional neural networks (CNN) have shown great results for object-detection tasks by learning texture and pattern-extraction filters. However, object-level interactions are harder to grasp without increasing the complexity of the architectures. On the other hand, Point Process models propose to solve the detection of the configuration of...