RoadNet-v2: A 10 ms Road Segmentation Using Spatial Sequence Layer
share
In automated driving systems (ADS) and advanced driver-assistance systems (ADAS), an efficient road segmentation module is required to present the drivable region and to build an occupancy grid for path planning components. The existing road algorithms build gigantic convolutional neural networks (CNNs) that are computationally expensive and time consuming. In this paper, we explore the usage of recurrent neural network (RNN) in image processing and propose an efficient network layer named spatial sequence. This layer is then applied to our new road segmentation network RoadNet-v2, which combines convolutional layers and spatial sequence layers. In the end, the network is trained and tested in KITTI road benchmark and Cityscapes dataset. We claim the proposed network achieves comparable accuracy to the existing road segmentation algorithms but much faster processing speed, 10 ms per frame.
READ FULL TEXT