Exploring Self-Attention for Visual Intersection Classification

Self-attention has recently emerged as a technique for capturing non-local contexts in robot vision. This study introduced a self-attention mechanism into an intersection recognition system to capture non-local contexts behind the scenes. This mechanism is effective in intersection classification because most parts of the local pattern (e.g., road edges, buildings, and sky) are similar; thus, the use of a non-local context (e.g., the angle between two diagonal corners around an intersection) would be effective. This study makes three major contributions to existing literature. First, we proposed a self-attention-based approach for intersection classification. Second, we integrated the self-attention-based classifier into a unified intersection classification framework to improve the overall recognition performance. Finally, experiments using the public KITTI dataset showed that the proposed self-attention-based system outperforms conventional recognition based on local patterns and recognition based on convolution operations.