Vehicular crowdsensing could enhance the traffic accident detection performance on freeways to the existing infrastructures by global task allocation. However, the uncertainty of traffic accidents and unknown of individual users bring difficulties for the task allocation. To address the problem, a global optimization method of task allocation with non-deterministic tasks and unknown users (GOTA-NU) is proposed. In the method, to reduce the influence of accident uncertainty, the accident risk is used to represent the sensing tasks of traffic accidents, and sensing task representation model is constructed by estimating the intrinsic temporal-spatial distribution of accident risk according to the wavelet transform and optimal transport theory. Meanwhile, to determine the users in future, a user estimation model is established according to macro statistical characteristics of traffic flow. Then the task allocation problem is transformed into a coverage problem for accident risk. A task allocation model is constructed by minimizing the user incentive cost with coverage level constraints of accident risk. And a greedy algorithm is proposed to solve it. To validate the proposed method, sensitivity experiments, robustness experiments, and comparison experiments are carried out on an open data source. The results show that the proposed method is efficient and reliable under different traffic conditions. The proposed method provides a reference for the long-term task allocation with non-deterministic tasks and unknown users.

Road travel time estimation (RTTE) is crucial for route planning, especially in urban connected vehicle environments. However, current methods struggle to extract the latent traffic semantics embedded in trajectory data, limiting their ability to provide fine-grained travel time predictions for individual road segments compared to overall routes. In this study, we introduce DeepTrafficSense, a comprehensive RTTE framework that takes into account the heterogeneity and temporal variability of road segment conditions. It integrates a heterogeneous dynamic graph (HDG) and a trajectory motion transformer (TMT) to model macro traffic dynamics and micro vehicle motion patterns. The HDG captures the temporal evolution of traffic states across road segments using asynchronous memory updates and adaptive feature aggregation, effectively addressing challenges such as irregular trajectory sampling and stale memory states. The TMT, designed for variable-length multivariate sequences, employs self- and cross-attention mechanisms to fuse endogenous (e.g., speed) and exogenous (e.g., acceleration) motion features. Additionally, a type-aware gating mechanism dynamically adjusts the contributions of the HDG and TMT, enhancing prediction accuracy for both intersections and regular road segments. Experiments on large-scale datasets from Chengdu and Xi’an demonstrate that DeepTrafficSense significantly outperforms state-of-the-art models in RTTE tasks.

Vehicular crowdsensing provides a new approach for freeway traffic accident detection. However, the uncertainty on traffic accidents and Mobile Users (MUs) brings great challenges for participant recruitment in constructing the deterministic representation of sensing tasks and estimating the participants. To address the challenges, a participant recruitment method for freeway traffic accident detection is proposed. In the method, to deal with the non-deterministic sensing tasks and MUs, the temporal-spatial distribution of accident risk is estimated by optimal transport theory to represent sensing tasks, and the probability distributions of MUs' trip distance and requested rewards are used to estimate MUs. Then the participant recruitment problem is converted into an optimal coverage problem for accident risk under the macro statistical characteristics of MUs. The participant recruitment model is established to determine the participants by maximizing the coverage rate of accident risk with the budget constraint. And a greedy heuristic strategy is used to solve the model. Simulation experiments are carried out to validate the proposed method. The results show the proposed method is effective and reliable in freeway traffic accident detection.

Using a deep learning method to detect buried pipes on ground penetrating radar (GPR) data is popular for its excellent performance potential. However, this needs a large amount of high-quality training data, which leads to time-consuming and labor-intensive data annotation work. Semi-supervised learning method provides a solution for the situation of small sample size. However, the original semi-supervised learning methods lack control over the quality of pseudo labels during model training. To address the problem, a new prior knowledge-guided semi-supervised deep learning method is proposed to improve the model performance under the small sample. In the method, a prior structure feature (PSF) is constructed to control the quality of pseudo label during semi-supervised learning. The structure response index (SRI) is designed to segment the structure subject out for eliminating disturbance from unstructured information. Then, the PSF is represented using histograms of oriented gradients (HOGs) and Fourier descriptors (FDs) to describe the structure’s edge orientation and shape. Based on the PSF, LightGBM is employed as a discriminator to screen out high-quality pseudo labels for model training. In the experiments, comparison experiments with some state-of-the-art supervised learning methods and semi-supervised learning methods were carried out to validate the method’s performance. The ablation studies on labeled dataset size, pseudo-label confidence threshold, and PSF were conducted to validate the effectiveness of the prior structural feature in controlling the pseudo label’s quality. The results show that the PSF is effective, and the proposed method outperforms the other methods.

Pavement segregation reduces pavement strength and causes early pavement damages. As a convenient and nondestructive subsurface detection technology, Ground Penetrating Radar (GPR) is widely applied to detect segregation by manual interpretation. This induces labor-intensive and time-consuming. And then the segregation is interpreted automatically by image processing technology based on GPR B-scan images. However, this ignores the intrinsic physical properties of segregation. To address the problem, a new method for automatically detecting segregation is proposed to interpret segregation by combining visual features with intrinsic physical properties of segregations. In the method, GPR data preprocessing is carried out to remove the subgrade layer disturbance and enhance GPR signal. And then the candidate segregation locating is put forward by segregation representation index and graph cuts algorithm according to segregation visual and statistical characteristics. Segregation visual features and intrinsic physical properties are represented for recognition according to gray level co-occurrence matrix and dielectric properties. The extreme gradient boosting is employed as a classifier to recognize segregation. In experiments, experiment samples are collected by 3D GPR testing vehicle and coring experiments, and comparison experiments are carried out to validate the proposed method by analysis on feature selection, segregation localization and segregation recognition. The results show the proposed method has a better performance than the other methods, such as multi-grained cascade forest, deepLab v3+ and SegNet. As a result, the proposed method will provide a new reference for automatic interpretation GPR application, especially, on the combining visual features with intrinsic physical properties.

Deep learning has achieved good performance in short-term traffic forecasting recently. However, the stochasticity and distribution imbalance are main characteristics to traffic flow, and these will bring the uncertainty and induce the network overfitting problem during deep learning. To deal with the problems, a new end-to-end hybrid deep learning network model, named M-B-LSTM, is proposed for short-term traffic flow forecasting in this paper. In the M-B-LSTM model, an online self-learning network is constructed as a data mapping layer to learn and equalize the traffic flow statistic distribution for reducing the effect of distribution imbalance and overfitting problem during network learning. Besides, the deep bidirectional long short-term memory network (DBLSTM) is introduced to reduce the uncertainty problem by forward and reverse contexts approximation process in the stochasticity reducing layer, and then the long short-term memory network (LSTM) is used to forecast the next traffic flow state in the forecasting layer. Furthermore, sufficient comparative experiments have been conducted and the results show the proposed model has better ability on solving uncertainty and overfitting problems than the state-of-art methods.