Efficient Lane Detection Technique Based on Lightweight Attention Deep Neural Network

Abstract

For self-driving vehicles, detecting lane lines in changeable scenarios is a fundamental yet challenging task. The rise of deep learning in recent years has contributed to the thriving of autonomous driving. However, existing methods of lane detection based on deep learning have high requirements on computing environment, so their applicability is further restricted. This paper proposed an improved attention deep neural network (DNN), a lightweight semantic segmentation architecture catering for efficient computation in low memory, which contains two branches worked in different resolution. The proposed network integrates fine details captured by local interaction of pixels at high resolution into global contexts at low resolution, computing dense feature maps for prediction task. Based on the attributes of disparate feature resolution characteristics, different attention mechanisms are adopted to guide the network to effectively exploit the model parameters. The proposed network achieves comparable results with state-of-the-art methods on two popular lane detection benchmarks (TuSimple and CULane), with faster calculation efficiency at 259 frames-per-second (FPS) on CULane dataset, and the total number of model parameters only requires 1.57 M. This study provides a practical and meaningful reference for the application of lane detection in memory constrained devices.

Evaluation of deep-learning based lane detection under low-light environments

“…The decoder is then embedded to obtain differentiable lanes. In [11], a lightweight semantic segmentation architecture is designed, which uses an improved attention deep neural network to achieve efficient computation with low memory. The network aggregates the fine details captured by local interaction of high-resolution pixels into a low-resolution global context, and computationally intensive feature maps are used for prediction tasks.…”

Lane Detection with Deep Learning: Methods and Datasets

“…There are also some works that focus on the computational speed and computational load of the model. A lightweight attention deep neural network with 4 modules and two branches is proposed in [28]. The input image goes through the global context embedding module, which encodes long-range contexts, and the explicit boundary regression model, which encodes low-level high-resolution feature maps in parallel after light downsampling.…”

“…Emerging graph neural networks have also been applied [10], and the processing of continuous images is also under study [4]. While some methods pursue precision, some works try to reduce the computational load [10,25,28,33] or use unsupervised learning [6]. In addition, there are also approaches that combine traditional computer vision methods with machine learning [5].…”

Vision-Based Branch Road Detection for Intersection Navigation in Unstructured Environment Using Multi-Task Network

“…Object detection methods using the bounding box are used for autonomous driving, and recent studies [12][13][14] propose improved methods to obtain robust results in a variety of weather and light conditions. Methods for improving detection accuracy [15] and computing efficiency [16] have been proposed. In addition, a method for recognizing roads in various environments (structured, unstructured, lane/ line-based, and curb) has been proposed [17].…”