Peter Roch

@inproceedings{talako-book-chapter,

title = {Positionierung induktiv geladener Fahrzeuge},

author = {Peter Roch and Bijan Shahbaz Nejad and Marcus Handte and Pedro José Marrón},

editor = {Heike Proff and Markus Clemens and Pedro José Marrón and Benedikt Schmülling},

doi = {10.1007/978-3-658-39979-5_5},

isbn = {978-3-658-39979-5},

year  = {2023},

date = {2023-05-05},

booktitle = {Induktive Taxiladung für den öffentlichen Raum: Technische und betriebswirtschaftliche Aspekte},

pages = {93--142},

abstract = {Ziel des TALAKO Projekts ist es, kabelloses Laden von Elektrofahrzeugen im öffentlichen Raum zu ermöglichen. Induktives Laden erfordert eine präzise Ausrichtung des Fahrzeugs, um einen effizienten Ladevorgang zu gewährleisten. Dabei hat die Ausrichtung des Fahrzeugs direkten Einfluss auf den Wirkungsgrad. Der Positionierungsvorgang kann für den Fahrer herausfordernd sein, da er den Versatz der Ladekomponenten ohne weitere Unterstützung nicht wahrnehmen kann. Daher umfasst die entwickelte Anlage neben der induktiven Ladeinfrastruktur selbst ebenfalls ein kamerabasiertes Fahrerassistenzsystem. Das Fahrerassistenzsystem wird dazu genutzt, anfahrende Fahrzeuge zu erkennen und den Fahrer beim Positionierungsvorgang zu unterstützen. Es besteht aus zwei Komponenten: einem kamerabasierten Positionierungssystem und einer Fahrerleitanwendung. Das Positionierungssystem nutzt Kamerabilder, um die Position von Fahrzeugen mit einer Genauigkeit von 5 cm zu berechnen. Daraus wird der Abstand zwischen Fahrzeug und Ladeplatte abgeleitet. Die Fahrerleitanwendung interpretiert die Positionsinformationen und generiert daraufhin geeignete Anweisungen für den Fahrer. Das Positionierungssystem basiert auf einem neuronalen Netz, welches die Reifen des Fahrzeugs erkennt. Da der Abstand zwischen den Reifen bekannt ist, kann daraus die Position und Rotation des Fahrzeugs errechnet werden. Untersuchungen haben ergeben, dass die Genauigkeit im Bereich von 5 cm liegt. Um das Positionierungssystem unabhängig vom Fahrzeugtyp und Installationsort zu betreiben, muss es entsprechend konfiguriert werden. Dazu muss das neuronale Netz trainiert und die Kameraausrichtung kalibriert werden. Das Training des neuronalen Netzes wird mit synthetisch generierten Bildern ergänzt, welche mit einem eigens entwickelten Bildgenerator produziert werden können. Die Kameraausrichtung wird mit einem speziellen Muster bestimmt, welches an verschiedenen Stellen auf dem Untergrund platziert wird. Da die realen Maße des Musters bekannt sind, lässt sich daraus die Geometrie des Installationsortes ableiten. Im Rahmen einer Nutzerstudie wurde untersucht, welche Bildschirmmodalität für die Fahrerleitanwendung unter den gegebenen Umständen optimal eingesetzt werden kann. Die Studie hat ergeben, dass Nutzer einen im Fahrzeug befindlichen Bildschirm für die Ausgabe von Anweisungen bevorzugen. Daher wurde die Fahrerleitanwendung durch eine mobile Anwendung realisiert. Diese zeigt dem Fahrer die Position des Fahrzeugs in Relation zur Ladestation an. Für die Darstellung der räumlichen Relationen wurden verschiedene Visualisierungen miteinander verglichen. Mit mehreren Visualisierungen sind die Nutzer in der Lage, das Fahrzeug in einem Toleranzbereich von 5 cm zu positionieren. Die meisten Nutzer bevorzugen jedoch eine Darstellung aus der Vogelperspektive. Die Kommunikation der beiden Komponenten wurde mittels Bluetooth Low Energy umgesetzt. Im Gegensatz zu anderen drahtlosen Kommunikationsmöglichkeiten, wie z. B. WLAN, bietet dies den Vorteil, dass Informationen ohne Verzögerung eines Verbindungsaufbaus an die mobile Anwendung gesendet werden können. Dadurch kann der Fahrer unmittelbar nach Ankunft an der Anlage die Positionierung verzögerungsfrei starten. Das Gesamtsystem wurde prototypisch bei einem Taxiunternehmen in Mülheim a. d. R. (Auto Stephany GmbH (2012) Auto Stephany GmbH – Taxi Dienstleistungen. Abgerufen am 04. 08. 2022 von https://taxi-stephany.de/) in Betrieb genommen und über mehrere Monate iterativ optimiert. Während dieser Zeit wurden wertvolle Erfahrungen gesammelt, die dazu beigetragen haben, dass sowohl das Positionierungssystem als auch die Fahrerleitanwendung stetig verbessert wurden. Nach Abschluss der Optimierungen konnte das entwickelte System erfolgreich als Bestandteil der Pilotanlage in Köln mit mehreren Ladeplätzen eingesetzt werden. Da die Pilotanlage in Köln im öffentlichen Raum betrieben wird, müssen die Persönlichkeitsrechte einzelner Personen beachtet werden. Eine explizite Einwilligung in die Datenverarbeitung durch die Betroffenen ist jedoch nicht praktikabel. Daher wurde eine automatisierte Verschleierung eingesetzt, welche personenbezogene Daten wie Kennzeichen und Gesichter aus den Kamerabildern entfernt, um eine Verarbeitung zu vermeiden.

},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}

@inproceedings{epic,

title = {Enhancing Privacy in Computer Vision Applications: An Emotion Preserving Approach to Obfuscate Faces},

author = {Bijan Shahbaz Nejad and Peter Roch and Marcus Handte and Pedro José Marrón},

editor = {George Bebis and Bo Li and Angela Yao and Yang Liu and Ye Duan and Manfred Lau and Rajiv Khadka and Ana Crisan and Remco Chang},

doi = {10.1007/978-3-031-20716-7_7},

isbn = {978-3-031-20716-7},

year  = {2022},

date = {2022-10-04},

urldate = {2022-10-04},

booktitle = {Advances in Visual Computing},

volume = {17},

pages = {80--90},

publisher = {Springer Nature Switzerland},

abstract = {Computer vision offers many techniques to facilitate the extraction of semantic information from images. If the images include persons, preservation of privacy in computer vision applications is challenging, but undoubtedly desired. A common technique to prevent exposure of identities is to cover peoples' faces with, for example, a black bar. Although emotions are crucial for reasoning in many applications, facial expressions may be covered, which hinders the recognition of actual emotions. Thus, recorded images containing obfuscated faces may be useless for further analysis and investigation. We introduce an approach that enables automatic detection and obfuscation of faces. To avoid privacy conflicts, we use synthetically generated faces for obfuscation. Furthermore, we reconstruct the facial expressions of the original face, adjust the color of the new face and seamlessly clone it to the original location. To evaluate our approach experimentally, we obfuscate faces from various datasets by applying blurring, pixelation and the proposed technique. To determine the success of obfuscation, we verify whether the original and the resulting face represent the same person using a state-of-the-art matching tool. Our approach successfully obfuscates faces in more than 97{%} of the cases. This performance is comparable to blurring, which scores around 96{%}, and even better than pixelation (76{%}). Moreover, we analyze how effectively emotions can be preserved when obfuscating the faces. For this, we utilize emotion recognizers to recognize the depicted emotions before and after obfuscation. Regardless of the recognizer, our approach preserves emotions more effectively than the other techniques while preserving a convincingly natural appearance.},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}

@inproceedings{guild,

title = {GUILD - A Generator for Usable Images in Large-Scale Datasets},

author = {Peter Roch and Bijan Shahbaz Nejad and Marcus Handte and Pedro José Marrón},

editor = {George Bebis and Bo Li and Angela Yao and Yang Liu and Ye Duan and Manfred Lau and Rajiv Khadka and Ana Crisan and Remco Chang},

doi = {10.1007/978-3-031-20716-7_19},

isbn = {978-3-031-20716-7},

year  = {2022},

date = {2022-10-04},

urldate = {2022-10-04},

booktitle = {Advances in Visual Computing},

volume = {17},

pages = {245--258},

publisher = {Springer Nature Switzerland},

abstract = {Large image datasets are important for many different aspects of computer vision. However, creating datasets containing thousands or millions of labeled images is time consuming. Instead of manual collection of a large dataset, we propose a framework for generating large-scale datasets synthetically. Our framework is capable of generating realistic looking images with varying environmental conditions, while automatically creating labels. To evaluate usefulness of such a dataset, we generate two datasets containing vehicle images. Afterwards, we use these images to train a neural network. We then compare detection accuracy to the same neural network trained with images of existing datasets. The experiments show that our generated datasets are well-suited to train neural networks and achieve comparable accuracy to existing datasets containing real photographs, while they are much faster to create.

},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}

@inproceedings{car-pose-estimation,

title = {Car Pose Estimation through Wheel Detection},

author = {Peter Roch and Bijan Shahbaz Nejad and Marcus Handte and Pedro José Marrón},

editor = {George Bebis and Vassilis Athitsos and Tong Yan and Manfred Lau and Frederick Li and Conglei Shi and Xiaoru Yuan and Christos Mousas and Gerd Bruder},

doi = {10.1007/978-3-030-90439-5_21},

isbn = {978-3-030-90439-5},

year  = {2021},

date = {2021-10-06},

urldate = {2021-10-06},

booktitle = {Advances in Visual Computing},

volume = {16},

pages = {265--277},

publisher = {Springer International Publishing},

abstract = {Car pose estimation is an essential part of different applications, including traffic surveillance, Augmented Reality (AR) guides or inductive charging assistance systems. For many systems, the accuracy of the determined pose is important. When displaying AR guides, a small estimation error can result in a different visualization, which will be directly visible to the user. Inductive charging assistance systems have to guide the driver as precise as possible, as small deviations in the alignment of the charging coils can decrease charging efficiency significantly. For accurate pose estimation, matches between image coordinates and 3d real-world points have to be determined. Since wheels are a common feature of cars, we use the wheelbase and rim radius to compute those real-world points. The matching image coordinates are obtained by three different approaches, namely the circular Hough-Transform, ellipse-detection and a neural network. To evaluate the presented algorithms, we perform different experiments: First, we compare their accuracy and time performance regarding wheel-detection in a subset of the images of The Comprehensive Cars (CompCars) dataset. Second, we capture images of a car at known positions, and run the algorithms on these images to estimate the pose of the car. Our experiments show that the neural network based approach is the best in terms of accuracy and speed. However, if training of a neural network is not feasible, both other approaches are accurate alternatives.},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}

@inproceedings{10.1007/978-3-030-90439-5_15,

title = {Evaluating User Interfaces for a Driver Guidance System to Support Stationary Wireless Charging of Electric Vehicles},

author = {Bijan Shahbaz Nejad and Peter Roch and Marcus Handte and Pedro José Marrón},

editor = {George Bebis and Vassilis Athitsos and Tong Yan and Manfred Lau and Frederick Li and Conglei Shi and Xiaoru Yuan and Christos Mousas and Gerd Bruder},

doi = {10.1007/978-3-030-90439-5_15},

isbn = {978-3-030-90439-5},

year  = {2021},

date = {2021-10-04},

urldate = {2021-10-04},

booktitle = {Advances in Visual Computing},

volume = {16},

pages = {183--196},

publisher = {Springer International Publishing},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}

@inproceedings{image-processing-pipeline-optimization,

title = {Systematic Optimization of Image Processing Pipelines Using GPUs},

author = {Peter Roch and Bijan Shahbaz Nejad and Marcus Handte and Pedro José Marrón},

editor = {George Bebis and Zhaozheng Yin and Edward Kim and Jan Bender and Kartic Subr and Bum Chul Kwon and Jian Zhao and Denis Kalkofen and George Baciu},

doi = {10.1007/978-3-030-64559-5_50},

isbn = {978-3-030-64559-5},

year  = {2020},

date = {2020-10-05},

booktitle = {Advances in Visual Computing},

volume = {15},

pages = {633--646},

publisher = {Springer International Publishing},

address = {Cham},

series = {Lecture Notes in Computer Science},

abstract = {Real-time computer vision systems require fast and efficient image processing pipelines. Experiments have shown that GPUs are highly suited for image processing operations, since many tasks can be processed in parallel. However, calling GPU-accelerated functions requires uploading the input parameters to the GPU's memory, calling the function itself, and downloading the result afterwards. In addition, since not all functions benefit from an increase in parallelism, many pipelines cannot be implemented exclusively using GPU functions. As a result, the optimization of pipelines requires a careful analysis of the achievable function speedup and the cost of copying data. In this paper, we first define a mathematical model to estimate the performance of an image processing pipeline. Thereafter, we present a number of micro-benchmarks gathered using OpenCV which we use to validate the model and which quantify the cost and benefits for different classes of functions. Our experiments show that comparing the function speedup without considering the time for copying can overestimate the achievable performance gain of GPU acceleration by a factor of two. Finally, we present a tool that analyzes the possible combinations of CPU and GPU function implementations for a given pipeline and computes the most efficient composition. By using the tool on their target hardware, developers can easily apply our model to optimize their application performance systematically.},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}

@inproceedings{driver-guidance-system,

title = {A Driver Guidance System to Support the Stationary Wireless Charging of Electric Vehicles},

author = {Bijan Shahbaz Nejad and Peter Roch and Marcus Handte and Pedro José Marrón},

editor = {George Bebis and Zhaozheng Yin and Edward Kim and Jan Bender and Kartic Subr and Chul Bum Kwon and Jian Zhao and Denis Kalkofen and George Baciu},

doi = {10.1007/978-3-030-64559-5_25},

isbn = {978-3-030-64559-5},

year  = {2020},

date = {2020-10-05},

booktitle = {Advances in Visual Computing},

volume = {15},

pages = {319--331},

publisher = {Springer International Publishing},

address = {Cham},

abstract = {Air pollution is a problem in many cities. Although it is possible to mitigate this problem by replacing combustion with electric engines, at the time of writing, electric vehicles are still a rarity in European cities. Reasons for not buying an electric vehicle are not only the high purchase costs but also the uncomfortable initiation of the charging process. A more convenient alternative is wireless charging, which is enabled by integrating an induction plate into the floor and installing a charging interface at the vehicle. To maximize efficiency, the vehicle’s charging interface must be positioned accurately above the induction plate which is integrated into the floor. Since the driver cannot perceive the region below the vehicle, it is difficult to precisely align the position of the charging interface by maneuvering the vehicle. In this paper, we first discuss the requirements for driver guidance systems that help drivers to accurately position their vehicle and thus, enables them to maximize the charging efficiency.  Thereafter, we present a prototypical implementation of such a system. To minimize the deployment cost for charging station operators, our prototype uses an inexpensive off-the-shelf camera system to localize the vehicles that are approaching the station. To simplify the retrofitting of existing vehicles, the prototype uses a smartphone app to generate navigation visualizations. To validate the approach, we present experiments indicating that, despite its low cost, the prototype can technically achieve the necessary precision.},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}

@inproceedings{9287891,

title = {Quantifying the Impact of the Physical Setup of Stereo Camera Systems on Distance Estimations},

author = {Alexander Julian Golkowski and Marcus Handte and Peter Roch and Pedro José Marrón},

doi = {10.1109/IRC.2020.00041},

year  = {2020},

date = {2020-01-01},

booktitle = {2020 Fourth IEEE International Conference on Robotic Computing (IRC)},

pages = {210-217},

abstract = {The ability to perceive the environment accuratelyis a core requirement for autonomous navigation. In the past,researchers and practitioners have explored a broad spectrumof sensors that can be used to detect obstacles or to recognizenavigation targets. Due to their low hardware cost and highfidelity, stereo camera systems are often considered to be aparticularly versatile sensing technology. Consequently, there hasbeen a lot of work on integrating them into mobile robots.However, the existing literature focuses on presenting theconcepts and algorithms used to implement the desired robotfunctions on top of a given camera setup. As a result, the rationaleand impact of choosing this camera setup are usually neitherdiscussed nor described. Thus, when designing the stereo camerasystem for a mobile robot, there is not much general guidancebeyond isolated setups that worked for a specific robot.To close the gap, this paper studies the impact of the physicalsetup of a stereo camera system in indoor environments. To dothis, we present the results of an experimental analysis in whichwe use a given software setup to estimate the distance to anobject while systematically changing the camera setup. Thereby,we vary the three main parameters of the physical camerasetup, namely the angle and distance between the cameras aswell as the field of view. Based on the results, we derive severalguidelines on how to choose the parameters for an application.},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}