Local refinement mapping system and method based on SLAM and semantic segmentation

The invention belongs to the field of computer vision, and discloses a local refinement mapping system and method based on SLAM and semantic segmentation, a binocular camera is used for video acquisition of images, a semantic segmentation module is built by a DANet model based on an FCN network framework, and the semantic segmentation module is built by a DANet model based on an FCN network framework. Processing the scene image by applying a channel attention module and a space attention module in the network model in the DANet model to realize target recognition and semantic segmentation; performing pose estimation on the feature points extracted from the semantically segmented target scene and the same feature points in other images through a front-end visual mileage module; the back-end optimization module selects key frames from the global camera pose, establishes global constraints by using the relationship of key frame quality inspection, and constructs a pose graph; and the closed-loop detection module