Test Methodology for Vision-Based ADAS Algorithms with an Automotive Camera-in-the-Loop

In order to correctly perceive its surroundings, advanced driver-assistance systems (ADAS) rely on the data quality of environment sensors, such as cameras, and on the data processing to distinguish multiple classes of traffic participants. Real test drives are important for their testing and validation, but certain test scenarios are difficult to be reproduced or automated, e.g. adverse weather conditions. Therefore, it is essential to bring this system to a controlled virtual environment so that it is possible to determine their correctness and performance under these circumstances before their release. For this reason, Hardware-in-the-Loop testing methods have been increasingly utilized in the industry, with which real hardware is connected to driving simulation software and deficiencies can be identified in a early development phase. This paper presents a test setup with a real automotive Camera-in-the-Loop and a testing method to evaluate a proprietary algorithm for multi-class object detection of an ADAS platform available on the market and validate the specifications described by its manufacturer.