Traffic environment perception is the foundation of autonomous driving, and road type recognition can provide support information for decision making of intelligent vehicle. In this paper, we adopt the strategy of integrating the Mask R-CNN model with two different convolutional neural networks (ResNet-101 and ResNet-50) respectively to perform the road type recognition and establish road type dataset containing pictures of expressway and urban main road based on the TT100K dataset. The experiment results indicate that Mask R-CNN with ResNet-50 can achieve the mAP of 96.30%, which is 17.60% higher than that of ResNet-101 and can realize the recognition of road type effectively. And the comparison between ResNet-50 and ResNet-101 demonstrates that for different recognition tasks with different amount of data, suitable depth network needs to be chosen carefully in order to achieve satisfying results. Key words: Mask R-CNN, road type, object recognition;

CICTP