Recent research has focused mostly on semantic segmentation using only RGB images. However, in adverse weather conditions, such as nights with low light or heavy rain obscuring most objects, the performance of RGB image-based models can significantly ...
Recent research has focused mostly on semantic segmentation using only RGB images. However, in adverse weather conditions, such as nights with low light or heavy rain obscuring most objects, the performance of RGB image-based models can significantly deteriorate. To solve this problem, semantic segmentation models based on the sensor fusion of long-wave infrared (LWIR) images and RGB images have been studied. However, those models encountered a problem of slow inference speed when embedded into edge devices. Therefore, this paper proposes an approach to improving the inference speed of the semantic segmentation model that fuses LWIR images and RGB images.