Cameras operating in the visual range of the electromagnetic spectrum are central to advanced driver assistance systems (ADAS). Front cameras, analyzing traffic, are often located behind the windshield to detect and classify objects.Thus, the area of the windshield within the camera’s field of view is a part of the optical system. Simple windshields consist of two curved glass surfaces connected by a thermoplastic interlayer. Due to defects present in the raw glass, as well as those introduced during the bending and lamination process, windshields will have optical aberrations. While optical quality may be suitable for human vision, it can fall short of what is needed for machine vision. In this article we investigate how the optical aberrations generated by laminated safety glass (LSG) influence the optical performance of a camera system and based on this, how the classification of image content by a convolutional neural network (CNN) is affected. A method for wavefront measurements of LSG samples is presented, which allows us to parameterize a linear optical model in Zernike Space. From this, we derive space-variant point spread functions (PSFs) and apply those to the dataset to simulate the windshield’s impact on the camera image. As a use case, a CNN was trained on the unmodified dataset and compared to the modified versions with the LSG models applied. We measured and modelled two different LSG samples, one with high and the other one with low optical quality. We compare the prediction accuracy of the classification with the unmodified data. The highquality sample had negligible effect on the overall classification accuracy, while the low-quality sample lowered the prediction accuracy by up to ten percentage points due to the optical aberrations.