Attention‐based multi‐scale feature fusion for free‐space detection

Free space detection is a very important task in road scene understanding. With the continued development of convolutional neural networks, free-space can be seen as class-specific semantic segmentation problem. In this paper, new encoding–decoding network structure-HRUnet designed, which always maintains input high-resolution images both encoding and decoding phases. It extracts multi-scale information from RGB continuously fuses them, finally achieves accurate spatial detection. addition, order to improve accuracy detection, attention mechanism module-spin proposed achieve interaction between channel dimensions when calculating attention, establish come relationship space, reduce loss feature information, further Experimental results show that structure outperforms current popular models terms balanced computational complexity accuracy.