Evaluating the Robustness of Object Detection in Autonomous Driving System
Xiang Bi, Hui Gao, He Chen, Pengqi Wang, Chenghao Ma · 2022 9th International Conference on Dependable Systems and Their Applications (DSA) · 2022
Autonomous vehicles are facing many challenges because of various types of scenarios with different conditions. Object detection is a computer vision task and the foundation of high-level tasks of autonomous driving operation, such as object tracking and path planning. Object detection model performance can vary due to various weather conditions and extreme scenarios. In this article, we propose a method to evaluate the robustness of object detection in autonomous vehicles with three metrics. The result of experiment shows that as the level of image corruptions increase, the performance of standard object detection trained on large-scale datasets prominently drop in our robustness metrics.