Combining Static Specular Flow and Highlight with Deep Features for Specular Surface Detection

To apply robot teaching to a factory with many mirror-polished parts, it is necessary to detect the mirror-like surface accurately. Deep models for mirror detection have been studied by designing mirror-specific features, e.g., contextual contrast and similarity. However, the mirror-polished parts, e.g., plastic molds, tend to have complex shapes and ambiguous boundaries, and thus existing mirror-specific deep features could not work well. To detect such complex mirror-like surfaces, we propose combining static specular flow and highlight, frequently appearing in specular surfaces, with deep model-based multi-level feature pyramids and adaptively integrating multiple feature maps, including mirror-specific ones. Through experiments with our original real-world plastic mold dataset, we show the effectiveness of the proposed method.