Schneider, Stefan-Alexander
Refine
Year of publication
Document Type
- conference proceeding (article) (24)
- 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 (32)
- nicht begutachtet (18)
Keywords
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.
Rob Harwood, Ansys global industry director, met with professor Stefan-Alexander Schneider from the University of Applied Sciences at Kempten in Bavaria, Germany. Professor Schneider runs one of the world’s only master’s courses on advanced driver assistance systems and autonomous vehicles. Their conversation revealed how the University of Applied Sciences is helping to develop the disruptive technologies of tomorrow and the engineers who will deliver them. They all discussed why simulation is critical for autonomous vehicles.
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 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.
Numerische Methoden
(2006)
Diese Einführung in die Numerische Mathematik behandelt die Themenbereiche Rechengenauigkeit, lineare Gleichungssysteme, Interpolation, Integration, Fouriertransformation, Nullstellenbestimmung sowie gewöhnliche und partielle Differentialgleichungen sehr anschaulich. Der Schwerpunkt liegt auf effizienten, rechnergestützten Lösungsansätzen, z. B. Wavelets, Splines und Mehrgitterverfahren. Viel Wert wird dabei auf aktuelle Anwendungsbeispiele aus dem Umfeld Computer Science gelegt, wie Bildverarbeitung, Computer-Graphik, Data Mining und Wettervorhersage. Historische Beispiele ergänzen die Darstellung. Dieses Lehrbuch eignet sich somit für Studierende der Informatik, Mathematik sowie Ingenieur- und Naturwissenschaften, die einen modernen Zugang zum Einsatz numerischer Methoden suchen.
Die Neuauflage wurde aktualisiert und Erfahrungen der täglichen Vorlesungspraxis eingearbeitet. Neue Anwendungsbereiche wurden mit aufgenommen, insbesondere aus der Softwareentwicklung für industrielle Anwendungen und aus dem Bereich des Internet. Zusätzlich wurde als Unterstützung eine Webseite eingerichtet, die es den Studierenden ermöglicht, interaktiv und mit Lösungshilfen die Übungsaufgaben des Buches selbstständig zu bearbeiten.
In this paper, we describe a new approach for synthetic image augmentation and its advantages in training Deep Neural Networks (DNNs) for object classification and localization. To address the need for a significant amount of data when training DNNs, for image-based ADAS functions, our method relies on virtually generated scenarios augmented via a physics-based camera model. The camera model implements various optical effects on ideal-synthetic images. For the scope of this paper, we illustrate the performance differences associated with the vignetting effect when training DNNs with and without image augmentation. We show that training on images altered by our camera vignetting model yield to a better performance than using ideal-synthetic images, additionally we illustrate the relationship between the network's performance results and the implemented effect (vignetting in this case). For a start, our results open the possibility for using camera models for training neural networks on synthetic data and pave the way toward further investigations on significant optical and image sensor effects to be modeled/implemented for performance enhancement during the training process. The approach is conducted and evaluated by training a DNN for car detection using the Karlsruhe Institute of Technology and Toyota Technological Institute at Chicago (KITTI) and Virtual KITTI (VKITTI) datasets.
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.
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)