Advanced driver assistance systems increase road safety and are an active research field in the automotive industry. In recent years, the complexity of the algorithm for advance driver assistance systems, e. g. by object detection instead of situation representation, and the complexity of the electronic control units (ECU), e. g. by the increased numbers of automotive certified processor systems, have increased strongly. During production development, this growth of complexity is typical the reason for increased development cost and can lead to delays of the product lunch. For development of the algorithm and the specification of the ECU, the knowledge of the requirements for an algorithm in terms of execution time and cost is therefore already of particular interest in the research and pre-development stage. This monograph addresses the simulation-based performance evaluation of possible ECU architectures for camera-based advanced driver assistance systems. For the performance evaluation of ECU architectures, a pedestrian detection system based on a sensor fusion approach of two different camera systems was selected as a test application. The complexity of the selected application is one of the most challenging use cases for such performance evaluation systems. The number of pedestrian and the changing background cause different execution times and demands the modeling of dynamic dataflow. One goal of the work is to develop a methodology for the integration of a system-level virtual prototype in the development process of advance driver assistance system. It will be shown, that by the usage of the development tool SystemCoDesigner a performance evaluation in terms of execution time and cost is possible before proof of concept and in addition a reduction of the development time of advance driver assistance systems of up to six months is possible due to the virtual prototyping on the electronic system-level. Based upon this, the main focus of this monograph is the determination of the simulation results accuracy of the system-level virtual prototype, and the development effort for the simulation-based performance evaluation of camera-based advanced driver assistance systems. For this purpose, the performance, in particular the throughput of frames, and the usage of different resources, i. e. ASIP, multi-core processor, FPGA, and a heterogeneous multi-processor system-on chip consisting of a dual-core processor and an FPGA, is evaluated For the performance evaluation, an actor-oriented model of the pedestrian detection system was developed, which represents the parallelization opportunities of the algorithm for different target devices. The introduced actor-oriented model consists of 22 actors. A method to calibrate the simulation model of a multi-core processor system is introduced in detail which allows an accurate prediction of the execution time by the system-level virtual prototype. For the calibration, a component-based determination of the necessary execution times is proposed. The time for execute a task on a processor is determined independent from the time for transfer data between processors. This approach allows a prediction of the execution time for parallel implementations of the advance driver assistance system before a parallel implementation is necessary. For the first time, the influence of compiler optimization and scheduler overhead is taken into account in the simulation of the execution time. By execution of a calibration script, the calibration of the system-level virtual prototype is finished within several seconds after the measurement results are available. In this monograph, the sources are presented which influences the accuracy of the system-level virtual prototype. A thorough experimental evaluation shows the high accuracy of the predicted execution time by the system-level virtual prototype for different test-scenarios. For 2 different test sequences with 500 test-scenarios, quality and fidelity of the simulation results for implementations on a multi-core processor, an ASIP, and an FPGA was evaluated. The average quality was between 0.801 and 0.973 and the fidelity between 91.6 % und 99.7%. In summary, the usage of a system-level virtual prototype might be one of the driving technologies during development of advance driver assistance system. The system-level virtual prototype not only allows the performance evaluation of future ECUs, it enables a product launch six months earlier and an automated parallel implementation for demonstration of the advance driver assistance system in the experimental car already in the research and pre-development stage.
Simulative Leistungsbewertung für Fahrerassistenzsysteme zur Fußgängerdetektion: Modellierung, Kalibrierung, Validierung
2014
163 Seiten, Bilder, Tabellen, Quellen
Hochschulschrift
Deutsch
Leistungsbewertung , Sensorfusion , Bearbeitungsanlage , Antriebssystem , Objekterkennung , Produktentwicklung , Verarbeitungsmaschine , Automobilindustrie , Entwicklungskosten , elektronisches Steuergerät , Monographie , Detektorsystem , Nachweiseinrichtung , Verkehrssicherheit , Fußgänger , Entwicklungsprozess , Elektronikgerät , Eichen (Abgleichen) , Compiler
Kamerabasierte Fußgängerdetektion
Springer Verlag | 2015
|Fahrerassistenzsysteme - Validierung von Hochband Sensorik
Kraftfahrwesen | 2016
|Fahrerassistenzsysteme - Validierung von Hochband Sensorik
Kraftfahrwesen | 2016
|Springer Verlag | 2019
|