Virtual Validation for Advanced Driver Assistance Systems

Road safety remains a critical global issue, as traffic accidents frequently occur, primarily due to human error. In response, Advanced Driver Assistance Systems (ADAS) have been developed to reduce accident risks. However, traditional validation methods, which rely on physical road tests, remain essential but are costly, time-consuming, and limited to a narrow range of use cases, failing to cover hazardous or particularly complex scenarios that, although occurring occasionally in the real world, are crucial for a complete and effective validation of monitoring systems. This thesis explores the application of virtual validation and synthetic data in testing automotive monitoring systems, with a focus on In-Cabin monitoring and Surround View camera systems. The first part of this thesis investigates the integration of a generative model into a virtual validation framework to generate driver animations for testing purposes. This model simplifies the process of generating realistic and diverse driver behaviors in virtual scenarios, which would otherwise require considerable effort and time to cover a wide range of driving scenarios. By simplifying the generation of these animations, the model enables faster, more efficient, and repeatable validation of In-Cabin monitoring systems, allowing for comprehensive testing across different situations without the need for manual frame-by-frame animation creation. The second part focuses on validating Surround View technology for ADAS. A virtual environment was developed to test Wide-FoV camera systems more effectively, addressing the limitations of traditional flat-screen testing methods. By overlaying real world image captures with virtual environment simulations, the study analyzes distortions and camera positioning errors, providing an evaluation of the system’s performance.