Semantic Segmentation Model for Road Scene Based on Encoder-Decoder Structure

摘要

Semantic segmentation as a pixel-wise segmentation task provides rich object information, which is an important research topic in robotic perception. It has been widely applied in many fields, such as autonomous driving and robot navigation. In the application of understanding road scene, the semantic segmentation model should accurately describe the appearance and shape of different categories of objects. In addition, the semantic segmentation model need to understand the spatial relationships between different categories. In order to improve the performance of semantic segmentation model for road scene, we present a model based on encoder-decoder structure with dilated convolution. We apply this model on the Cityscapes dataset and compare it with other classical models. To assess performance, we rely on the standard Jaccard Index IoU (Intersection over Union) and mIoU (mean Intersection over Union). The experimental results verify that this model can effectively improve the performance of semantic segmentation and meet the requirements for road scene.