Advanced driving assistance systems (ADASs) play a vital role in the safety of transportation. The detection of lane markings is a very important part of ADASs. For the safety of autonomous vehicles and vehicles driven by human drivers, accurate detection results are necessary. In this study, the authors propose a novel algorithm based on lane structural analysis and convolutional neural networks (CNNs) for lane marking detection. First, a pre‐processing stage is used to remove the pavement that constitutes the background of the lane markings. Next, a set of local waveforms from local images is used to generate a region of interest and a CNN classifier is employed to detect lane marking candidates. Finally, a lane geometry analysis stage determines whether or not the candidate is a part of a lane marking. The major contributions of this study can be summarised as follows. First, they propose a novel method to describe a road using waveforms. Second, they analyse the local and global characteristics of the road geometry to detect the lane markings. Third, they provide an effective method to obtain training data for the proposed machine learning method. Experimental results demonstrate that the proposed method outperforms conventional methods.

Track geometry monitoring is essential for track maintenance. Dedicated track inspection vehicles are scheduled to measure track geometry irregularities throughout the railway network, so cannot inspect each line frequently. It is desirable to inspect geometry much higher frequently using in-service trains. One possible way is estimating track geometry from vehicle–body accelerations because the accelerators can be easily installed in vehicle–body. However, inverting track geometry from vehicle–body acceleration is a pending issue. Up to now, most research has focused on vehicle dynamics modelling and simulation. In this paper, we solve the problem using deep learning method and realistic measurement data. The training and test data are the inspection data acquired by comprehensive inspection trains from three main high-speed railways in China. The proposed AM–CNN–GRU model combines an Attention Mechanism (AM), a Convolutional Neural Network (CNN) and a Gated Recurrent Unit (GRU). The model’s inputs are vertical and lateral accelerations, and vehicle speed. The outputs are two vertical track irregularities, with wavelengths of 3–42m and 3–120 m, respectively. We evaluated the model by comparing the estimated irregularities with the actual measurements. We discussed different models, high-speed lines, sub-rail infrastructures and speed to validate the model effectiveness.

Classification of traffic signs with partial occlusions is important for traffic sign maintenance and inventory systems. It is also important to help drivers identify possible traffic signs in time. Motivated by human cognitive processes in identifying an occluded sign, a novel structure is designed to explicitly handle occluded samples in this paper. Occlusion maps are analyzed for possible occluded signs, and a new occlusion descriptor is proposed to distinguish occluded signs from negative samples. A series of tests shows that the developed method could effectively handle samples with partial occlusions and thus reduce the missed detections caused by occlusions. The developed method could also be easily used for any other object detection.

To improve the quality of track maintenance work, it is a desire to estimate vehicle dynamic behavior from track geometry irregularities. This paper proposes a deep learning model to predict vehicle responses (e.g., vertical wheel-rail forces, wheel unloading rate, and car body vertical acceleration) using deep learning techniques. In the proposed CA-CNN-MUSE model, convolutional neural networks (CNNs) are used to learn features of track irregularities, and multiscale self-attention mechanisms (MUSE) are employed to capture the long-term and short-term trends of sequences. Coordinate attention (CA) is introduced into CNN to focus on important interchannel relationships and important spatial mileage points. The experiments were performed on a multibody simulation model of the vehicle system and the measured data of the actual high-speed line. The results show that the CA-CNN-MUSE has high prediction accuracy for vertical vehicle responses and fast computation speed. The predicted time-domain waveforms and power spectral densities (PSDs) agree well with the actual vehicle responses. The main features of the lateral vehicle responses can also be captured by the proposed method, yet the results are not as good as the vertical ones.