Noise Reduction of Segmented Images by Spatio-Temporal Morphological Operations

摘要

To realize more effective and reasonable systems for robot navigation, this paper proposes a novel visual navigation scheme for robot navigation that strongly depends on the result of semantic segmentation. The visual navigation scheme showed good performance in some actual scenes. However, the current implementation does not work well if noise regions emerge on a movable area after semantic segmentation. To solve this problem, this paper proposes a novel scheme for noise reduction of segmented images based on three-dimensional morphological operations using spatio-temporal images. Experimental results showed that the number of noise regions could be reduced to zero while keeping the false negative value at 1. In addition, the proposed scheme could remove noise regions even though a smaller number of frames was used for filtering than in the case of the spatio-temporal mode filter previously proposed.