Development of a method to predict image sensor remaining useful lifetime

As autonomous vehicles become increasingly prevalent on the roads, ensuring the safety and reliability has never been more critical. One of the key components in achieving this safety is the monitor the performance of the camera sensor. To address these concerns, this thesis investigates an aspect of sensor reliability: predicting the remaining useful life (RUL) of the image sensor particularly AR0220AT On Semiconductor sensor. The research focuses on the utilization of thermal shock cycling as an aging methodology and reliability test, simulating the sensor's aging process and to study its effects on object detection confidence, image quality, and dead pixel count. This approach aimed to replicate real-world conditions and assess how these factors influence camera performance. Additionally, synthetic degradation studies were performed using the KITTI dataset to evaluate potential impacts on detection and segmentation. Although the study did not achieve direct RUL prediction due to the minimal performance impact observed from the aging process, it provided invaluable insights into sensor evaluation. The findings highlighted that while aging had a relatively minor effect compared to severe weather conditions like rain and fog, the research also established critical methodologies for assessing camera performance and image quality.
This thesis contributes significantly to the field by outlining essential steps for evaluating sensor RUL and performance. It lays the groundwork for future research by detailing how alternative aging techniques could be applied to predict RUL more accurately, ultimately enhancing the safety and reliability of autonomous vehicle systems.