Lens Flare Aware Detector In Autonomous Driving

Autonomous driving is crucial for reducing traffic accidents, and object detection plays a key role. However, challenges exist in complex driving conditions. This paper focuses on the study of object detection in the presence of lens flare, which is one optical artifact mentioned in the IEEE P2020 Automotive Imaging White Paper. To the best of the author’s knowledge, this is the first paper analyzing the impact of lens flare on object detection in autonomous driving tasks and proposing a lens flare adaptation method based on Bayesian reasoning theory to optimize existing object detection models, achieving higher average precision. To conduct our experiments, we first divided an existing autonomous driving dataset (BDD100K) into daytime and nighttime subsets. We then added lens flares at the location of sky or light sources to synthesize a dataset with lens flare effects in autonomous driving tasks. The Mean Square Error (MSE) value of the detected object image is regarded as the degree of lens flare. Then based on the dataset, we construct Log-Likelihood Ratio (LLR) curves for different degrees of lens flare and use these curves to adjust the confidence score (which can represent the probability of correctly detecting the object) of the detected objects of the detector. Based on our experiments, it has been demonstrated that our proposed method is effective in improving the Average Precision (AP) metric in object detection for almost all classes in the BDD100K dataset. Furthermore, this method only requires simple modifications based on the detection results of the existing object detection models, making it easier to deploy on existing devices.