Windshield Quality Control for Perfect Image Recognition
Dr. Olaf Thiele
The future Advanced Driver Assistant Systems (ADAS) of higher levels need to analyze highway driving conditions at longer distant ranges. With longer focal lengths of the ADAS lenses the quality of all optical components become more relevant. The glass windshield design influences the imaging performance of the entire camera-based ADAS system drastically. Therefore, the analysis of the windshield´s optical quality in combination with the lens is necessary. The lenses are designed for longer distances of 60 m and more. Especially for the viewing area of the camera field in the windshield, the camera working ranges at a hyperfocal distance require new, extended test criteria.
The current optical quality of windshields is specified by standards like ECE R43 and ANSI SAE Z26. One relevant value for the ADAS camera image quality is the optical distortion strength or optical power, expressed in terms of Millidiopter (mdpt). According to the EU-standard the optical distortion at the driver´s view must be less than 145mdpt, but most of the automotive OEMs define individually lower standards. Especially for the ADAS camera view field area the standard is often a third or less of this value. But new developments show, the optical distortion is not sufficient to characterize impacts to the ADAS camera. The qualitative optical analysis of the SFR (Spatial Frequency Response) or MTF (Modulation Transfer Function) is required to evaluate the transmission optics or image quality.
Figure 1: Camera-based system for the distortion and SFR analysis of windshields
The LaVision´s camera-based measurement system provides the analysis of the optical distortion and the SFR of windshields together. The full-field imaging technique accurately measures the optical power of local distortions in milli-diopters with the highest spatial resolution. The system uses a structured background pattern and the digital image correlation (DIC) technique of images with and without the windshield for this application. The second quality value the system can analyze is the spatial frequency response of the windshield. The SFR is measured pointwise following the ISO12233 standard by using slanted edges at the target.
Figure 2: Horizontal and vertical distortion map of an ADAS camera area at a windshield
In Figure 2 an example of a diopter map of a ADAS camera area at a tilted windshield is shown. At the horizontal distortion the vertical float lines from the production are visible, even in this small scale with very high spatial resolution. In this case some effects from the black frit at the rim of the ADAS field can also be detected. The vertical distortion values are due to the tilting of the windshield higher.
Comparing these diopter results with SFR measurements evidences the discrepancy between these two values. Vertical profile of the vertical distortion along the center of the ADAS camera view area at the windshield is shown in Figure 3. The diopter values decrease from the top of the area to the bottom continuously. The evolution of the vertical SFR ratio at the same positions by a vertical scan shows drastically changes between 90% and below 10% at this windshield sample. The SFR ratio is 100% when there is no impact to the optical contrast after placing the windshield in front of the
measurement system. In this example the optical distortion and the spatial frequency response got a completely different structure and indicate, that there is no direct correlation between these two optical evaluation values. This is generally the fact for bended windshields with multiple glass layers and interlayer polymer PVB-foils.
Figure 3: Comparison of the vertical distortion to the vertical SFR at the same position
Figure 4: Camera example images without the windshield (Ref), at positions with low (Pos1) and high (Pos2) SFR ratio values
For visualizing the effect of high and low SFR ratio values raw camera images of the measurement are shown in Figure 4. The reference image Ref shows the slanted edges of the target without a windshield with a perfectly focused analysis system. When introducing the windshield at the relevant inclination angle into the view field of the camera-based LaVision´s system especially at the position Pos1 on the windshield an extreme blurring of the tilted horizontal edges in the vertical direction is visible. This leads to a SFR ratio value below 10% here. At the second position Pos2 on the windshield with a SFR ratio value of nearly 90% only a minimal blurring is affecting the image quality.
Figure 5: Optional setup with a number of SFR targets in the camera view with specified viewing angles
Using the scanning method along defined lines across the windshield can give precisely information of the local optical quality changes within the material and is very helpful especially during the development phase of new windshields tailored to ADAS cameras. The impact of different viewing angles can be analyzed by placing a number of SFR targets in the view field of the camera (see Figure 5). The different positions can now be evaluated in a single image without scanning of the LaVision system. This is a typical setup for windshield quality control in production lines.
The ongoing research shows a correlation between AI image recognition of object detection and classification in ADAS applications and the spatial frequency response SFR of the windshields.
Compared to the normal classification with the optical power the relationship is much stronger. This emphasizes the necessity to qualify windshields by the SFR for camera-based ADAS applications.
The LaVision glass analysis system is capable of these SFR evaluations using the customer ADAS lenses for the optical quality control of windshields. The ability to analyze two optical-transmissive measured values increases the benefits for a wide range of users and provides the investigation of influencing variables of ADAS- camera view fields and to further safeguard camera-assisted, future autonomous driving at levels 3 and higher.
Product Manager Automotive LaVision
