With the growth of deep learning technology, there are many elaborate object detection models being developed for safe autonomous driving. However, a common problem is that the training data is often biased toward normal daytime which leads to high un...
With the growth of deep learning technology, there are many elaborate object detection models being developed for safe autonomous driving. However, a common problem is that the training data is often biased toward normal daytime which leads to high uncertainty in the predictions on adverse weather conditions that were not included in the training data. Therefore, in this paper, we developed a robust model for bad weather conditions by utilizing mixture density network to estimate the uncertainty of the deep learning model’s predictions. Our method showed better performance than original models in fog, rain, and nighttime environments.