This book is a collection of proceedings of the talks given at the 2015 annual joint workshop of Fraunhofer IOSB and the Vision and Fusion Laboratory (IES) by the doctoral students of both institutions. The topics of individual contributions range from computer vision, optical metrology, and world modelling to data fusion and human-machine interaction.

This work proposes a Multibody Structure from Motion (MSfM) algorithm for moving object reconstruction that incorporates instance-aware semantic segmentation and multiple view geometry methods. The MSfM pipeline tracks two-dimensional object shapes on pixel level to determine object specific feature correspondences, in order to reconstruct 3D object shapes as well as 3D object motion trajectories.

Mit der zunehmenden Menge an Bilddaten im Videoüberwachungssektor wächst die Chance, Straftaten besser aufklären zu können. Allerdings ist dafür ein immenser Aufwand für die Auswertung der Bilder erforderlich, die oft nicht mehr vollständig ohne Computerunterstützung durch Personen gesichtet werden können. Diese Arbeit umfasst Methoden und Verbesserungen auf Basis neuartiger Personenrepräsentationen für die Detektion, Verfolgung und erscheinungsbasierte Wiedererkennung von Personen. With the increasing amount of image data in the video surveillance sector, the opportunity to better solve crimes is growing. However, this requires an immense effort for the exploitation of the images, which often can no longer be viewed completely without computer assistance by persons. This work comprises methods and improvements based on novel person representations for the detection, tracking and appearance-based recognition of persons.

The Proceeding of the annual joint workshop of the Fraunhofer IOSB and the Vision and FusionLaboratory (IES) 2018 of the KIT contain technical reports of the PhD-stundents on the status of theirresearch. The discussed topics ranging from computer vision and opticalmetrology to network security and machine learning.This volume provides a comprehensive and up-to-date overview of the research program of the IESLaboratory and the Fraunhofer IOSB.

Ihrer Arbeit in der Originalsprache: Steigende Variantenvielfalt der Produkte und hohe Flexibilitätsanforderungen der Prozesse resultieren in zunehmender Komplexität der Automobilproduktion. Neben der Beherrschbarkeit dieser Herausforderungen ist zur Gewährleistung der Qualitätsanforderungen sowie einer effizienten Qualitätsregelung eine intensive Mensch-System-Kollaboration in der Montagelinie erforderlich. Dazu liefert diese Arbeit fünf methodisch-technische Beiträge zur Regelung der Produktionsqualität in der Fahrzeugmontage. Increasing variety of products and high flexibility requirements of production processes result in increasing complexity of automobile production. In addition to mastering these challenges, intensive human-system collaboration in the assembly line is required to guarantee quality requirements and efficient quality control. This dissertation therefore provides methodological and technical contributions to control production quality on the basis of five interrelated sections.

Zur Erschließung allgemeiner Prinzipien für das Themenfeld der Bewegungsplanung für vollautomatisches Fahren wird eine intuitive Problemformulierung als Euler-Lagrange-Modell aufgestellt und zur globalen Optimierung in ein korrespondierendes Hidden-Markov-Modell umgewandelt. Geometrische und topologische Betrachtungen führen zu einer probabilistischen Umgebungsmodellierung in Kombination mit dem C²-Modell und resultieren in universellen Schlussfolgerungen über die Struktur von Verkehrsgeschehen. In order to establish general principles in the topic of motion planning for fully-automated driving, an intuitive problem statement in the form of an Euler–Lagrange Model is derived and transformed into a corresponding Hidden Markov Model for global optimization. Geometric and topologic considerations lead to a probabilistic environment modelling in combination with the C² model and result in general conclusions about the structure of traffic situations.

Für die Mensch-Roboter-Interaktion wird in dieser Arbeit eine Methode zur Überwachung der komplexen, dynamischen Roboterumgebung vorgestellt. Die Roboterbewegung wird basierend auf dem Konzept der modellprädiktiven Regelung unter Berücksichtigung der detektierten Hindernisse und der stattfindenden Kontakte des Roboters mit seiner Umgebung geregelt, um Kollisionen zu vermeiden und angemessen auf Kontakte zu reagieren. Die Ansätze werden auf einem mobilen Manipulator validiert. For human-robot-interaction, this work proposes a method for monitoring the complex, dynamic environment of the robot. The robot motion is controlled based on the concept of nonlinear model predictive control. The controller considers the detected obstacles and intended or unintended contacts of the robot with its environment so that undesired collisions are avoided and an adequate reaction to contacts is achieved. The proposed approaches are analyzed on a mobile manipulator.

This book proposes to solve the low-resolution (LR) facial analysis problem with 3D face super-resolution (FSR). A complete processing chain is presented towards effective 3D FSR in real world. To deal with the extreme challenges of incorporating 3D modeling under the ill-posed LR condition, a novel workflow coupling automatic localization of 2D facial feature points and 3D shape reconstruction is developed, leading to a robust pipeline for pose-invariant hallucination of the 3D facial texture.

This book summarizes presentations delivered at the 2017 annual joint workshop of the Fraunhofer Institute of Optronics, System Technologies and Image Exploitation (IOSB) and the Vision and Fusion Laboratory (IES) of the Institute for Anthropomatics, Karlsruhe Institute of Technology (KIT). At this meeting, doctoral students from both institutions reported on the status of their research and discussed topics ranging from computer vision to network security and machine learning.

This work proposes a feature-based probabilistic data association and tracking approach (FBPDATA) for multi-object tracking. FBPDATA is based on re-identification and tracking of individual video image points (feature points) and aims at solving the problems of partial, split (fragmented), bloated or missed detections, which are due to sensory or algorithmic restrictions, limited field of view of the sensors, as well as occlusion situations.