A Novel Transparent Object Detection Approach Based on Boundary Optimization

Detecting transparent objects using vision sensors has always been a challenge in the industry. In this paper, a novel algorithm of semantic segmentation based on the edge information of transparent objects is proposed. We first use the input RGB image for semantic segmentation to obtain the segmentation images of transparent objects. We predict changes in surface pixel information and perform boundary detection. Based on the results obtained in the previous stages, we introduce a boundary optimization module to achieve image segmentation of transparent objects. In addition, the semantic segmentation method for transparent objects integrates the Efficient Transformer module and performs boundary optimization on the segmentation results to obtain more refined segmentation results. The experiment has demonstrated the effectiveness and accuracy of this method.