Refine
Year of publication
Document Type
- conference proceeding (article) (23)
- Article (10)
- conference proceeding (presentation) (8)
- Other (3)
- Part of a Book (2)
- Book (1)
- Doctoral Thesis (1)
- Moving Images (1)
- Report (1)
Publication reviewed
- begutachtet (31)
- nicht begutachtet (18)
Keywords
Infrastructure-cooperated autonomous driving systems are attracting attention as a method for promoting the practical application of highly functional autonomous driving. We focused on the part where recognition processing on the infrastructure side can be advanced, which is not possible with in-vehicle processing. Using a fixed-point camera and recognition of the situation behind a vehicle or object in which multiple cameras are linked are such examples. In this paper, we selected a difficult situation such as a curved road, focused on the scene where the vehicle is running while deforming its shape, and examined a method of accurately recognizing the vehicle using a fixed-point camera. It is a study of the criteria for dividing the vehicle shape class. Recognition of the general vehicle class of autonomous driving also needs to identify unknown objects and non-vehicles. In his article, we have excluded the identification of unknown objects and focused on recognizing known vehicles using deep learning. Consider six different vehicle shapes on curved roads. We investigated the impact of vehicle shape class integration and performance, and found that the integration of the two classes reduced the number of vehicle shape classes and increased recognition accuracy.
In autonomous driving, detecting vehicles together with their parts, such as a license plate is important. Many methods with using deep learning detect the license plate based on number recognition. However, there is an idea that the method using deep learning is difficult to use for autonomous driving because of the complexity in realizing deterministic verification. Therefore, development of a method that does not use deep learning(DL) has become important again. Although the authors have made the world's best performance in 2018 for Caltech data with using DL, this concept has now turned to another research without using DL. The CT5L method is the latest type, that includes techniques of the continuity of vertical and horizontal black-and-white pixel values inside the plate, unique Hough transform, only vertical and horizontal lines are detected, the top five in the order of the number of votes to ensure good performance. In this paper, a method to determine the threshold value for binarizing input by machine learning is proposed, and good results are obtained. The detection rate is improved by about 20 points in percent as compared to the fixed case. It achieves the best performance among the conventional fixed threshold method, Otsu's method, and the conventional method of JavaANPR.
Phase noise (PN) is one of the most significant impairments adversely affecting the detection performance of frequency-modulated continuous wave (FMCW) radar systems. Due to the rapid advance of advanced driver assistance systems (ADAS), virtual testing and the evaluation of highlyautomated driving (HAD) functions became indispensable. In this work, the impact of PN on the performance of automotive radar sensors is demonstrated on HAD functions in a virtual driving simulator. Therefore, a PN model initially developed for static objects is applied to dynamic scenarios including moving objects. By implementing a real world scenario in the virtual environment the influence of PN on the detection performance of the radar sensor is demonstrated. The virtual test scenario is implemented using the CarMaker test driving software, which is commonly accepted as an accurate and reliable tool by the automotive industry. The radar sensor model including PN is implemented as a functional mock-up unit (FMU) using the standardized functional mock-up interface (FMI) 2.0 and the open simulation interface (OSI) 3.0.0. Finally, the radar FMU model simulations are compared with hardware measurements.
In this paper, we derive intermediate frequency (IF) level analytical formulation of radio frequency (RF) group delay for automotive frequency-modulated continuous-wave (FMCW) radar waveform under quasi-static approximation. To the best of our knowledge, this paper is the first to develop and simulate an IF-level analytical form ulation of RF group delay, including random and deterministic variation for the FMCW radar waveform. Theoretical limitation for the tolerable RF group delay can be derived based on the proposed model. We demonstrated the impact of RF group delay on the FMCW radar sensor's range spectrum in dynamic virtual traffic scenarios. The proposed model is integrated into a virtual FMCW radar sensor model implemented as a functional mock-up unit (FMU) using the standardized interfaces functional mock-up interface (FMI) 2 .0 and the open simulation interface (OSI) 3. 0. 0. A virtual test scenario is implemented in an industry-standard simulation tool, CarMaker, to demonstrate the effect.
There is a method for deterministically detecting a license plate, which is one of the object detections in autonomous driving. It is difficult to determine a luminance threshold value used for binarization. Therefore, the method of predicting the threshold value by machine learning from the combination of surrounding luminances has been improved. First, we constructed an augmentation that extends the objective variable from the structure of the labeled data, and increased the number of the original data by about 50 times. Next, we devised a new method to prevent overestimation of the cross-validation method. After Augmentation, we developed a Leave A Group Out Cross-Validation method that separates training data and test data in groups. By combining the performance improvement by the configured Augmentation and the improvement by the conventional SMOTE, a detection rate of 0.92 was achieved. As a result, the autonomous driving module has been strengthened by one step.
Bewege mich!
(2019)
In the Automotive industry and especially in the ADAS domain, functions like “Vehicle Detection”, “Lane Detection” undergo a very costly and time-consuming validation process before their final deployment in the vehicle.
After the completion of the development process, image based detection algorithms usually rely on huge data sets of previously recorded data for performance testing and validation. Though vision data sets like “KITTI Vision Benchmark Dataset” and others are currently available for public use, there still lies numerous requirements that need
to be satisfied and steps that need to be followed in order to pave the way for a proper and meaningful use of the recorded data sets in the scope of image based function testing and validation. Using the publicly available recorded data may be in some cases a good starting point but as we all know sooner or later we will need a more
customized/personalized recorded data sets that capture more precise and detailed specifications like the camera’s technical specifications or even its mounting position in the car. Furthermore, depending on the image based function under investigation, recorded data should also reflect certain driving scenarios in specific environmental conditions (rain, snow, fog, at sun rise, at daytime, at night …) or specific driving parameters like, speed,acceleration, grip, car
orientation, position in lane, etc. that that may be too hard to safety due to safety, financial restrictions or even time
limitations.
Sparse grids are a recently introduced new technique for discretizing partial differential equations having a very favorable complexity in the number of unknowns for higher dimensional problems. Therefore, sparse grids are especially attractive for instationary equations when time is treated as an additional dimension. The paper will introduce the sparse grid finite element technique and the sparse grid combination technique which can be interpreted as a multivariate extrapolation method. The conceps are closely related to the multilevel principle so that multigrid methods and multilevel preconditioning strategies are the natural solvers. Thus the overall solution process has optimal complexity. Furthermore, the combination technique is easily parallelizable and applicable to nonlinear problems, like the Richardson equation. Besides an introduction of the algorithms with their basic analysis we will present numerical tests for a suite of characteristic model problems.
In deep learning, in order to improve learning performance, preprocessing and ingenuity to combine a plurality of discriminators are performed. It can be inferred that it has elements exceeding the set of learning. Therefore, a configuration to combine multiple recognition elements with low loss will be studied. The advance category classification method is expected to narrow the scope of learning in the next stage. Combining elements specialized for FalsePositive/FalseNegative removal after the positive/negative determination is considered to be effective if the accuracy of the subsequent stage is high. We conducted a license plate recognition experiment by combining these and achieved the best performance for Caltech data.
Deep Learning-Based Multi-scale Multi-object Detection and Classification for Autonomous Driving
(2019)
Autonomous driving vehicles need to perceive their immediate environment in order to detect other traffic participants such as vehicles or pedestrians. Vision based functionality using camera images have been widely investigated because of the low sensor price and the detailed information they provide. Conventional computer vision techniques are based on hand-engineered features. Due to the very complex environmental conditions this limited feature representations fail to uniquely identify a specific object. Thanks to the rapid development of processing power (especially GPUs), advanced software frameworks and the availability of large image datasets, Convolutional Neural Networks (CNN) have distinguished themselves by scoring the best on populthis information, the boundingar object detection benchmarks in the research community. Using deep architectures of CNN with many layers, they are able to extract both low-level and high-level features from images by skipping the feature design procedures of conventional computer vision approaches. In this work, an end-to-end learning pipeline for multi-object detection based on one existing CNN architecture, namely Single Shot MultiBox Detector (SSD) [1], with real-time capability, is first reviewed. The SSD detector predicts the object’s position based on feature maps of different resolution together with a default set of bounding boxes. Using the SSD architecture as a starting point, this work focuses on training a single CNN to achieve high detection accuracy for vehicles and pedestrians computed in real time. Since vehicles and pedestrians have different sizes, shapes and poses, independent NNs are normally trained to perform the two detection tasks. It is thus very challenging to train one NN to learn the multi-scale detection ability. The contribution of this work can be summarized as follows:
A detailed investigation on different public datasets (e.g., KITTI [2], Caltech [3] and Udacity [4] datasets). The datasets provide annotated images from real world traffic scenarios containing objects of vehicles and pedestrians.
A data augmentation and weighting scheme is proposed to tackle the problem of class imbalance in the datasets to enable the training for both classes in a balanced manner.
Specific default bounding box design for small objects and further data augmentation techniques to balance the number of objects in different scales.
Extended SSD+ and SSD2 architectures are proposed in order to improve the detection performance and keeping the computational requirements low.
In the early phase of new vehicle system developments, it is crucial to fully define and optimize working system and functional architectures. Architecture definition and validation in turn requires a quick and accurate evaluation of a system‟s overall performance. Modeling and simulating a complete vehicle system, however, is complex and in many cases was either technically not achievable or simply has been omitted within the development process. It is the utmost challenge in system modeling and simulation to realistically reflect interaction of various electrical, mechanical, thermal, and software elements as attributed to individual system modules and their relations. State-of-the-art tools meanwhile bear this capability. In this paper we present an approach how they may effectively and efficiently be incorporated into a car system development process. To accomplish this target, we „virtualize‟ all system entities while defining and reflecting all relevant system aspects. Our proposed development flow allows simulating, evaluating, and validating complete vehicle systems and their behavior. The proposed flow will sustainably change car system development processes.
Trajectory Modelling for Autonomous Driving: Investigating the Artificial Potential Field Method
(2024)
Although the focus of autonomous driving is on maximizing safety and efficiency, comfort and familiarity will play a key role in the adoption of autonomous driving. Therefore, it is important to develop algorithms that can mimic human driving skills and adapt to individual driving styles. The potential field method (PFM) is an obstacle avoidance algorithm for autonomous driving that uses a repulsive potential field, as a environment model, to navigate the vehicle to the lowest risk potential. In this paper, the PFM is used in a overtake scenario at high speed, to test the impact of using prediction when calculating the ideal yaw rate. Analysis is done on how the potential field can be used for lane keeping while following a car and then for overtaking it. A driving simulator is used to record human driving data and compare it with automated driving using a PFM as is proposed by [3], with modifications to enable future prediction.
The following paper points out the key role of IT in the future of car development. At the moment a fundamental change in the structure of automotive IT organizations can be observed. The fact that software update cycle in automotive, about 1 year, in comparison with Apple, Google or Tesla is too much. The entertainment industry is constantly proceeding ahead much faster than the automotive industry. On top of this, new emerging platforms like Apple CarPlay and Android Auto are providing the look and the feel of a mobile phone regarding the control of the car. The vehicle itself is getting more and more as an “ultimate mobile application or app”. This shows the need of speeding up the Time-to-Market of new innovations in automotive industry.
The structure of IT departments has to support these process. No wonder that CIOs of car manufacturers are looking for new structures in their IT departments that enable faster cycle update for automotive applications taking in consideration safety and security requirements.
This only represents a particular interest, as for Apple and Google, we can see that Google has already a fleet of 23 self-driving cars in place which has already autonomously driven more than one million miles with only 12 accidents on public roads and Apple is said to work under the project name "Titan" on its own electric car.
Another important aspect is the software running in the car itself, e.g. the software that “fuses” data from sensors into a comprehensible form: objects have to be accurately located in the environment model of the socalled ego vehicle as a basis for decisions making either by the driver himself or even by the software that can determine within a fraction of a second what the car is going to do. High definition maps also play a very important role in enabling autonomous driving, being developed and maintained by companies such as Nokia HERE, with accuracy of only a few centimeters are thought to be of strategic importance for Advanced Driver Assistance Systems and Self Driving Cars.
“We’re the engine room of the system,” says Mr. Ristevski, vice president of reality capture and processing for former Nokia’s mapping unit named HERE. To be independent from Apple and Google maps and with that from possible competitors, it is said to be the main reason why the German premium car manufacturer Audi, BMW and Daimler bought the online map service for about € 2.5 bn. This is only the first step in the restructuring of the automotive industry.
The goal of the presented work is to develop an automotive radar sensor behavioral model to outline closed loop interaction of driver and vehicle in a synthetic environment. For this interaction open simulation interface (OSI) and functional mock-up interface (FMI) standards are used. This paper describes the architecture overview, working concept of FMI, OSI and different radar functions like radar channel and digital signal processing (DSP) in detail. The description of the scenario for the verification of radar behavioral model results and development of highly automated driving (HAD) functions for the closed loop simulation is elaborated in this paper.
Der allgemeine technologische Fortschritt, insbesondere in der Halbleitertechnik, beschert allen Industriezweigen bisher ungeahnte Entwicklungsschübe, die ganz allgemein auch auf die Software-Entwicklung ausstrahlen und sich somit auch auf die Software-Entwicklungsmethoden auswirken. Die Entwicklungsmethoden für Steuergeräte-Software, befinden sich daher in dem Übergang von manueller C-Code Programmierung hin zu einem graphischen Entwurf. Diese Entwurfsmöglichkeiten bringen viele Vorteile mit sich, wie z.B. Verringerung der Komplexität, Verkürzung der Software-Entwicklungszeit, Verbesserung der Software-Qualität und Automatisierung des Software-Produktionsprozesses.
Allerdings wirft dieser Abstraktionsschritt eine wesentliche Frage auf: Wie lässt sich überprüfen und sicherstellen, dass die eingesetzten Software-Entwicklungswerkezeuge für die Transformation vom graphischen Modell zum generierten C-Code und darüber hinaus zum Objektcode fehlerfrei arbeiten? Typische Methoden, um das notwendige Vertrauen in diese Transformationen zu erhalten sind: Betriebsbewährtheit, Audit des Entwicklungsprozesses des Werkzeuges sowie das Validieren der Entwicklungswerkzeuge. Insbesondere beim Validieren der Entwicklungswerkzeuge fehlte ein Leitfaden wie ein solcher Validierungsprozess aussehen könnte und in welchem Umfang - Stichwort Testtiefe - ein Entwicklungswerkzeug zu testen wäre.
In diesem Beitrag wird gezeigt, wie diese Lücke methodisch geschlossen werden konnte: Die Anforderungen an einem Validierungsprozess wurden ganz allgemein für Entwicklungswerkzeuge hergeleitet und entsprechende Validierungssuite zur Qualifizierung von C-Codegeneratoren und Target-Compilern umgesetzt und angewendet. Die Ergebnisse fließen in die entsprechenden Entwicklungsprozesse in Form von Modellierungsrichtlinien ein. Dieses Vorgehen setzt damit den Stand der Technik.
Die Herleitungsbasis für die Anforderungen an eine Validierungssuite bildeten die Anforderungen an Software-Entwicklungswerkzeuge (Das Entwicklungswerkzeug soll die "Eigenschaft X" haben). Dazu wurden einschlägige Normen und Standards, wie DIN EN 61508, RTCA DO-178B, IEC 60880, PTB-Softwareprüfstelle und MOD EDF Std 00-55, sorgfältig analysiert und entsprechend kategorisiert. Dieser Anforderungskatalog wurde dann in einen in sich stimmigen Anforderungskatalog für eine Validierungssuite (Die Validierungssuite soll validieren, dass das Entwicklungswerkzeug die "Eigenschaft X" hat) umformuliert und veröffentlicht und können beim TÜV Nord bezogen werden.
Der Validierungsprozess ist zweistufig vorgesehen. In einem ersten Schrott wird ein Gutachten für die prinzipielle Eignung der Modellierungssprache eingeholt. Diese Vorprüfung der Eingabeelemente eines Entwicklungswerkzeuges dient dazu, das Entwicklungswerkzeug hinsichtlich all jener Anforderungen zu prüfen, deren Erfüllung von einer weitgehend automatisierten Validierungssuite nicht geprüft werden können. Auf diese Weise kann auch frühzeitig sichergestellt werden, ob grundsätzliche Abweichungen zu den Anforderungen einer Qualifizierung eines Entwicklungswerkzeuges im Wege stehen und ermöglicht somit eine frühzeitige Behebung. Die sich daran anschließende Hauptprüfung des Entwicklungswerkzeuges durch eine Validierungssuite prüft das Entwicklungswerkzeug ausführungsbasiert um die notwendige Testtiefe zu erreichen.
Im Ausblick wird diskutiert, ob und wie die Anforderungen zur prinzipiellen Eignung einer Modellierungssprache beim Entwurf neuer Modellierungssprachen für sicherheitsrelevante Entwicklungsprojekte an dem Beispiel der Modellierungssprache Modelica wegweisend sein können.