Simulation-based test methods with an automotive camera-in-the-loop for automated driving algorithms

Testing automated driving systems is challenging due to the numerous and unexpected scenarios these can en-counter in road traffic. In order to reduce the effort and costs of real test drives, different simulation-based test methods for environment sensors have been developed. The purpose is to enable the hardware integration of sensors in the early development phase of environment perception systems for au-tomated driving applications. Thus, the test results are also subject to the uncertainties and inaccuracies of the real sensors rather than the ideal sensor models. In this work, two different methods for injecting synthetic image data into camera sensors are investigated: Over-the-Air (OTA) and Direct Data Injection (DDI). The goal is to verify to what extent the method for data injection degrades the image quality and, thus, influences the performance of the YOLOv4 algorithm for object detection. It is shown that both methods have their advantages and limitations: DDI achieves better image quality with real and synthetic data after image injection; OTA could present itself as a cost-effective solution for specific test scenarios.