Gradient Map Based Lane Detection Using CNN and RNN

While lane detection in complex driving environments is challenging for traditional computer vision methods, researchers have proposed the use of neural networks to address the problem. Most work in the literature uses full-colour images as the input to the network. In this paper, we show that an edge-based gradient map input can help neural networks in terms of improved accuracy, shorter processing time and training time, especially for small neural networks that fit on low-power consumption platforms. We show that, in comparison to RGB images, gradient map based convolutional neural networks can achieve better accuracy at different scales, and a compressed gradient map network can achieve up to 3.6 times speedup on the inference time while keeping a similar performance. In addition, we show that a gradient map input can also be used for recurrent neural networks to improve lane detection in obscured situations.