Synthetic aperture radar (SAR) and optical satellite image registration is a field that developed in the last decades and gave rise to a great number of approaches. The registration process is composed of several steps: feature definition, feature comparison and optimization of a geometric transformation between the images. Feature definition can be done using simple traditional filtering or more complex deep learning (DL) methods. In this paper, two traditional approaches and a DL approach are compared. One can then wonder if the complexity of DL is worth to address the registration task. The aim of this paper is to quantitatively compare approaches rooted in distinct methodological areas on two common datasets with different resolutions. The comparison suggests that, although more complex, the DL approach is more precise than traditional methods.

Multi-view object detection is a fundamental problem in computer vision. Current approaches generally require an explicit partition between different views with or without sharing descriptors. We present a novel boosting based learning approach which automatically learns a multi-view detector without using intra-class sub-categorization based on prior knowledge. To avoid multiplying the false alarm rate by the number of classifiers, which happens on the classical approach where one classifier per view is considered, we build a single cascade of weak classifiers which contains an implicit hierarchical structure. In details, a partition of positive samples is automatically computed in order to build an adequate weak classifier based on one specific descriptor per subset. By adapting iteratively the number of descriptors at each stage, the so-defined hierarchical structure enables both a precise modelling and an efficient sharing of descriptors between views. Experimental results demonstrate the relevance and efficiency of this new approach.

Monitoring vegetation dynamics is a very important task for biodiversity conservation, especially in the Global Change context where many natural habitats are threatened with extinction or reduced to small areas. Remote sensing (aerial or spatial) has played a key role in this monitoring for years. With space imagery, the main limitation is actually due to the metric resolution which is not well adapted for estimating the processes of changes occurring in areas where the number species, visible only in few places, is shrinking, like in the heathland habitat in Fontainebleau Forest case. For aerial imagery, the main problem is the cost of a specific mission or the availability of the plane or the pilot. We propose in this article another solution: the use of a home-made Unmanned Vertical Take Off and Landing Aerial System, which is a good compromise between the spatial and aerial solutions, and test the validity and the robustness of the tools developed for the management of natural habitat on a specific area of the Fontainebleau Forest.

Lidar waveforms are 1-D signals representing a train of echoes caused by reflections at different targets. Modeling these echoes with the appropriate parametric function is useful to retrieve information about the physical characteristics of the targets. This paper presents a new probabilistic model based upon a marked point process which reconstructs the echoes from recorded discrete waveforms as a sequence of parametric curves. Such an approach allows to fit each mode of a waveform with the most suitable function and to deal with both, symmetric and asymmetric, echoes. The model takes into account a data term, which measures the coherence between the models and the waveforms, and a regularization term, which introduces prior knowledge on the reconstructed signal. The exploration of the associated configuration space is performed by a reversible jump Markov chain Monte Carlo (RJMCMC) sampler coupled with simulated annealing. Experiments with different kinds of lidar signals, especially from urban scenes, show the high potential of the proposed approach. To further demonstrate the advantages of the suggested method, actual laser scans are classified and the results are reported.

Training object detectors aims at choosing specific visual attributes which are efficient and optimal for each learned object. This paper presents a new process which achieves this goal by putting all families of descriptors one wants to consider in a pool of descriptors and by letting the algorithm build a cascade with the most efficient descriptors by introducing management of very large features pools. On the one hand, the selection of a specific family of descriptors for a given application implies a deep experience of the operator on the algorithm behaviour. On the other hand, physical constraints such as computer time and memory requirements prevent us from using all available descriptors. We present here a solution which considers several families of descriptors as a pool of descriptors and builds a cascade with the most efficient descriptors. The idea developed here consists in beginning to build a cascade with one type of descriptors and then introducing new kinds of descriptors when the current descriptor family does not bring enough differentiating information anymore. In this scope, four families of descriptors are studied here: Histogram Distance on Haar Region (HDHR), Edge Orientation Histograms (EOH), Histogram Orientation Gradient (HOG) and Gabor filters. Evaluation on public data sets shows the importance of complementary features, since performances of state of the art methods are improved.

This paper deals with the estimation of an elevation model using a pair of synthetic aperture radar (SAR) images and an optical image in semiurban areas. The proposed method is based on a Markovian regularization of an elevation field defined on a region adjacency graph (RAG). This RAG is obtained by oversegmenting the optical image. The support for elevation hypotheses is given by the structural matching of features extracted from both SAR images. The regularization model takes into account discontinuities of buildings thanks to an implicit edge process. Starting from a good initialization, optimization is obtained through an iterated conditional mode algorithm.

This paper proposes a contribution to the adjustment problem of airborne laser scanner strips. Based the comparison of each laser strips with a photogrammetric derived Digital Surface Model (DSM), and on the modeling of the discrepancies between these two data sets, the algorithm produces an homogeneous 3D deformation field. The measurements of these deformations is performed using a modified Hough transform and we search for the maximum of the accumulator whereupon we assign a probability of correctness. Depending of this probability, we estimate a global affine transform over the cloud of points. For transitivity properties, all strips are registered with regard to this DSM and therefore they all perfectly fit on overlapping areas. This fine registration is of importance for many applications, especially when fusionning photogrammetric data and lidar data for feature extraction. More generally, this method can be applied to any other sorts of 3D data and provides an efficient algorithm for matching 3D datasets

This paper explores a generic approach to predict the output accuracy of an algorithm without running it, by a careful examination of the local context. Such a performance prediction will allow one to qualify the appropriateness of an algorithm to treat images with given properties (contrast, resolution, noise, richness in details, contours or textures, etc.) resulting either from experimental acquisition conditions or from a specific type of scene. We have to answer the following question: a context c being given at any site, what will be the performance? In our experiments, c is described by three contextual variables: Gabor components, entropy and signal noise ratio. As initially proposed in the related work, the prediction function is determined from training using a logistic regression model. This technique is illustrated on aerial infrared images for two types of algorithm: edge detection and displacement estimation.

Accurate 3D surface models of dense urban areas are essential for a variety of applications, such as cartography, urban planning and monitoring mobile communications, etc. Since manual surface reconstruction is very costly and time-consuming, the development of automated algorithms is of great importance. While most of existing algorithms focus on surface reconstruction either in rural or sub-urban areas, we present an approach dealing with dense urban scenes. The approach utilizes different image-derived cues, like multiview stereo and color information, as well as the general scene knowledge, formulated in data-driven reasoning and geometric constraints. Another important feature of our approach is simultaneous processing of 2D and 3D data. Our approach begins with two independent tasks: stereo reconstruction using multiple views and region-based image segmentation, resulting in generation disparity and segmentation maps, respectively. Then, the information derived from the both maps is utilized for generation of a dense elevation map, through robust verification of planar surface approximations for the detected regions and imposition of geometric constraints. The approach has been successfully tested on complex residential and industrial scenes.

This paper presents a new method for the automatic registration of SPOT images and digitised maps. Direct matching of high level features (urban areas and crossroads) is performed using an hypothesis generation and propagation scheme. Results are presented for the registration of SPOT images and maps covering the same scene, as well as for the retrieval of partial maps into a large SPOT image.

We present a method to generate three-dimensional building hypotheses starting from sparse features, hypothesized building corners, extracted from multiple views of the scene. This technique relies on knowledge about the imaging geometry and acquisition parameters to provide rigorous geometric constraints for the matching process. The effectiveness of this approach is demonstrated using complex aerial imagery taken with highly oblique views containing buildings with flat or peaked roofs.>