Volker Steinhage

Extracting Buildings from Aerial Images Using Hierarchical Aggregation in 2D and 3D

We propose a model-based approach to automated 3D extraction of buildings from aerial images. We focus on a reconstruction strategy that is not restricted to a small class of buildings. Therefore, we employ a generic modeling approach which relies on the well-defined combination of building part models. Building parts are classified by their roof type. Starting from low-level image features we combine data-driven and model-driven processes within a multilevel aggregation hierarchy, thereby using a tight coupling of 2D image and 3D object modeling and processing, ending up in complex 3D building estimations of shape and location. Due to the explicit representation of well-defined processing states in terms of model-based 2D and 3D descriptions at all levels of modeling and data aggregation, our approach reveals a great potential for reliable building extraction.

Identification of Africanized honey bees through wing morphometrics: two fast and efficient procedures*

Currently available morphometric and genetic techniques that can accurately identify Africanized honey bees are both costly and time consuming. We tested two new morphometric techniques (ABIS — Automatic Bee Identification System and geometric morphometrics analysis) on samples consisting of digital images of five worker forewings per colony. These were collected from 394 colonies of Africanized bees from all over Brazil and from colonies of African bees, Apis mellifera scutellata (n = 14), and European bees, A. m. ligustica (n = 10), A. m. mellifera (n = 15), and A. m. carnica (n=15) from the Ruttner collection in Oberursel, Germany (preserved specimens). Both methods required less than five minutes per sample, giving more than 99% correct identifications. There was just one misidentification (based on geometric morphometrics analysis) of Africanized bees compared with European subspecies, which would be the principal concern in newly-colonized areas, such as the southern USA. These new techniques are inexpensive, fast and precise.ZusammenfassungDie Afrikanisierten Honigbienen sind unter den verschiedenen Unterarten und Rassengruppen der Honigbiene (Apis mellifera L.) in den Neotropen und den Nachbarregionen am meisten respektiert und gefürchtet, insbesondere da sie in neue Gebiete einwandern. Die Identifizierung der Afrikanisierten Bienen ist in diesen Regionen für die Bewirtschaftung der Bienenvölker daher unverzichtbar. Sie ermöglicht die Bestimmung ihres Verbreitungsgebiets und ihrer Ausbreitungsgeschwindigkeit, dies ist sowohl für die Imker als auch für die damit befassten Regierungseinrichtungen von Bedeutung.Wir benutzten zwei kürzlich entwickelte morphometrische Techniken (ABIS — Automatic Bee Identification System und die Geometrische Morphometrische Analyse), um Proben aus jeweils fünf rechten Vörderflügeln pro Volk zu analysieren (Tab. I). Beide dieser Methoden benötigten in einem Vergleich von 394 über ganz Brasilien verteilten Völkern weniger als 5 Minuten pro Volk und erreichten eine mehr als 99% korrekte Identifizierung. Diese ergaben 14 Völker von A. m. scutellata, 10 Völker von A. m. ligustica, 15 Völker von A. m. mellifera und 15 Völker von A. m. carnica (Tab. II und III). Mit ABIS können einzelne Bienen bestimmt werden, während die Geometrische Morphometrische Analyse eine auf jeweils 5 Flügeln beruhende Identifikationen auf Kolonieebene durchführt. Die meisten der Fehleinordnungen fanden zwischen Afrikanisierten und Afrikanischen Bienen sowie zwischen den europäischen Unterarten statt. Nur eines der Afrikanisierten Bienenvölker wurde irrtümlich als eine europäische Unterart eingeordnet, dies ist die Fehlerart die insbesondere innerhalb von neubesiedelten Gebieten wie den Südstaaten der USA von Bedeutung wäre. Die erreichten Fortschritte in Computertechnologie, statistischen Analysen und Bilderkennungssoftware sowie die verbesserten Informationen über die relevanten Messgrößenbereiche und die höhere Genauigkeit und größere Geschwindigkeit der Messungen selbst machen es nun möglich, Afrikanisierte Bienen ausschließlich anhand von Digitalaufnahmen der Vörderflügel in Minutenschnelle zu identifizieren.

Performance of histogram descriptors for the classification of 3D laser range data in urban environments

The selection of suitable features and their parameters for the classification of three-dimensional laser range data is a crucial issue for high-quality results. In this paper we compare the performance of different histogram descriptors and their parameters on three urban datasets recorded with various sensors-sweeping SICK lasers, tilting SICK lasers and a Velodyne 3D laser range scanner. These descriptors are 1D, 2D, and 3D histograms capturing the distribution of normals or points around a query point. We also propose a novel histogram descriptor, which relies on the spectral values in different scales. We argue that choosing a larger support radius and a z-axis based global reference frame/axis can boost the performance of all kinds of investigated classification models significantly. The 3D histograms relying on the point distribution, normal orientations, or spectral values, turned out to be the best choice for the classification in urban environments.

Models for photogrammetric building reconstruction

Abstract The paper discusses the modeling necessary for recovering man made objects—in this case buildings—in complex scenes from digital imagery. The approach addresses all levels of image analysis for deriving semantically meaningful descriptions of the scene from the image, via the geometrical/physical model of the objects and their counterparts in the image. The central link between raster image and scene are network-like organized aspects of parts of the objects. This is achieved by generically modeling the objects using parametrized volume primitives together with the application-specific constraints, which seems to be adequate for many types of buildings. The paper sketches the various interrelationships between the different models and their use for feature extraction, hypothesis generation, and verification.

Laser-based segment classification using a mixture of bag-of-words

In this paper, we propose a segment-based object detection approach using laser range data. Our detection approach is built up of three stages: First, a hierarchical segmentation approach generates a hierarchy of coarse-to-fine segments to reduce the impact of over- and under-segmentation in later stages. Next, we employ a learned mixture model to classify all segments. The model combines multiple softmax regression classifiers learned on specific bag-of-word representations using different parameterizations of a descriptor. In the final stage, we filter irrelevant and duplicate detections using a greedy method in consideration of the segment hierarchy. We experimentally evaluate our approach on recently published real-world datasets to detect pedestrians, cars, and cyclists.

Learning to hash logistic regression for fast 3D scan point classification

Segmenting range data into semantic categories has become a more and more active field of research in robotics. In this paper, we advocate to view this task as a problem of fast, large-scale retrieval. Intuitively, given a dataset of millions of labeled scan points and their neighborhoods, we simply search for similar points in the datasets and use the labels of the retrieved ones to predict the labels of a novel point using some local prediction model such as majority vote or logistic regression. However, actually carrying this out requires highly efficient ways of (1) storing millions of scan points in memory and (2) quickly finding similar scan points to a target scan point. In this paper, we propose to address both issues by employing Weiss et al.s recent spectral hashing. It represents each item in a database by a compact binary code that is constructed so that similar items will have similar binary code words. In turn, similar neighbors have codes within a small Hamming distance of the code for the query. Then, we learn a logistic regression model locally over all points with the same binary code word. Our experiments on real world 3D scans show that the resulting approach, called spectrally hashed logistic regression, can be ultra fast at prediction time and outperforms state-of-the art approaches such as logistic regression and nearest neighbor.

Person tracking in three-dimensional laser range data with explicit occlusion adaption

This paper presents an approach to exploit the richer information of sensor data provided by 3d laser rangefinders for the purpose of person tracking. Introduced is a method to adapt the observation model of a particle filter, to identify partial and full occlusions of a person, to determine the amount of occlusion behind an obstacle, and the occluding obstacle itself. This is done by tracing rays from positions near the person to the sensor and determining whether the ray hits an obstacle. The laser range data is represented using a voxel grid, which facilitates efficient retrieval and data reduction. As our experiments show, our proposed tracking approach is able to reliably keep track of a person in real-time, even when only partially visible, when moving in uneven terrain, or when the person passes closely another person of different size.

Pattern Recognition Combining Feature- and Pixel-Based Classification Within a Real World Application

Computer aided systems based on image analysis have become popular in zoological systematics in the recent years. For insects in particular, the difficult taxonomy and the lack of experts greatly hampers studies on conservation and ecology. We have developed a system for the automated identification of bee species which employs image analysis to classify bee forewings. The wings of the bees show a clear venation within a transparent surrounding. Our prototypical system [6] employs a semi automatic extraction of this venation by applying a line following procedure after interactively setting remarkable points of interest. Since both image distortions as well as random genetic mutations incorporate a high amount of uncertainty, the stage of feature extraction is left to an experienced taxonomical expert. However, due to the worldwide lack of experts in bee species we are confronted with the task of substituting the experts knowledge by a completely automated feature extraction. The remaining uncertainty in the location of some critical points then has to be dealt with additional features compensating the lack of information. Our new approach of combining both symbolic and suitable preprocessed iconic features together with either a Support Vector Machine or a nonlinear variant of dicriminant analysis as classifier enables a completely automated system with excellent performance.

LeafNet: A computer vision system for automatic plant species identification

Abstract Aims Taxon identification is an important step in many plant ecological studies. Its efficiency and reproducibility might greatly benefit from partly automating this task. Image-based identification systems exist, but mostly rely on hand-crafted algorithms to extract sets of features chosen a priori to identify species of selected taxa. In consequence, such systems are restricted to these taxa and additionally require involving experts that provide taxonomical knowledge for developing such customized systems. The aim of this study was to develop a deep learning system to learn discriminative features from leaf images along with a classifier for species identification of plants. By comparing our results with customized systems like LeafSnap we can show that learning the features by a convolutional neural network (CNN) can provide better feature representation for leaf images compared to hand-crafted features. Methods We developed LeafNet, a CNN-based plant identification system. For evaluation, we utilized the publicly available LeafSnap, Flavia and Foliage datasets. Results Evaluating the recognition accuracies of LeafNet on the LeafSnap, Flavia and Foliage datasets reveals a better performance of LeafNet compared to hand-crafted customized systems. Conclusions Given the overall species diversity of plants, the goal of a complete automatisation of visual plant species identification is unlikely to be met solely by continually gathering assemblies of customized, specialized and hand-crafted (and therefore expensive) identification systems. Deep Learning CNN approaches offer a self-learning state-of-the-art alternative that allows adaption to different taxa just by presenting new training data instead of developing new software systems.

Automated 3D reconstruction of grape cluster architecture from sensor data for efficient phenotyping

Graphical abstractDisplay Omitted Fully automated and detailed 3D reconstruction and phenotyping of grape clusters.Generative model of grape clusters to guide 3D reconstruction and phenotyping.Design and prototypic implementation of the complete processing chain.Qualitative and quantitative evaluation.Discussion of potential of generalization to other plants and with other sensors. We propose an approach to fully-automated and sensor-based 3D reconstruction of grape cluster architecture followed by a precise, objective, and reproducible derivation of phenotypic traits. Current approaches to sensor-based phenotyping often show interactive processing steps and analyze only those parts of a plant that can be sensed by the given sensor system. Our approach employs an explicit component-based model of the architecture of grape clusters, i.e., the interconnectivity of a grape clusters components, the geometry of the components, and the structural and geometrical constraints of their mutual connections. Based on this model, our approach can derive in a fully automated way complete 3D reconstructions of sensed grape clusters even for cases of partial occlusions in the process of sensor data acquisition. Given a complete 3D reconstruction of a grape cluster, we can derive on the one hand well known phenotypic traits of grape clusters. On the other hand, this approach facilitates measuring and evaluating new phenotypic traits. Therefore, our approach is of interest for monitoring and yield estimations in vineyards as well as for grapevine breeders. We developed and implemented our approach within a grapevine phenotyping project. First evaluations of reconstruction results and derived phenotypic traits show a potential of this approach for automated high-throughput phenotyping. We discuss the opportunities to apply our approach to other plants and with other sensor systems.