To address insufficient contextual information and partial loss of spatial detail information in semantic segmentation of road images, a real-time segmentation method is proposed based on LinkNet. Firstly, a new attention mechanism is constructed in the encoding to capture the location and channel dependence of road images to increase the extraction capability of target features. Then, an atrous spatial pyramid pooling is introduced in the central region to capture richer multi-scale features without affecting image resolution. The experimental results on the general database show that the proposed method achieves 64.78% MIoU on the Cityscapes dataset, which is 5.01% higher in comparison with LinkNet. And it can significantly improve the visual effect of fine target objects and boundary segmentation, and the segmentation accuracy has been greatly improved.