Road hazards can lead to dangerous accidents and endanger the safety of pedestrians. Frequent and thorough road inspection is required to maintain road safety. This paper proposes an improved U-Net model that combines gray-scale images and depth images and uses a data statistics method based on a road depth map to eliminate hazard-free data automatically and reduce the computational complexity involved in hazard detection. Experiments showed that the proposed model, based on the improved U-Net-based pavement hazard recognition and extraction algorithm, could smoothly and efficiently extract pavement cracks and deformation hazards in complex scenes with noise interference and produce results with strong robustness. Comprehensive indicators such as global recognition accuracy rate A (Accuracy), precision rate P (Precision), recall rate R (Recall), evaluation index F1 (F -Measure), and Mean Intersection over Union (MIoU) were used to assess the effectiveness of the proposed model in comparison with existing hazard detection models; the proposed model greatly outperformed the models in all the indicators. The proposed model can provide a significant reference for subsequent pavement repair work and be used to improve road safety.

In order to master the structure state of subway shield tunnel, reduce the diseases and avoid the potential safety hazards, the lateral convergence deformation of subway shield tunnels should be predicted. For the non-stationarity small sample size of subway shield tunnel lateral convergence deformation value, the existing prediction models have poor performance in terms of accuracy and stability. In this paper, a lateral convergence model of subway shield tunnel based on Kalman algorithm is constructed based on Kalman filtering theory. The model is efficient, adaptive and robust, and can accurately predict the lateral convergence deformation of subway shield tunnel. Taking the horizontal diameter of 200 rings shield segment in the interval section of a subway tunnel as an example, it is proved that the residuals of the Kalman prediction model are small, the residual distribution conforms to the normal distribution, and the prediction effect is great. The model is suitable for the prediction of more than 5 periods of data, and the prediction accuracy of the model is improved with the increase of the number of data periods. In addition, the Kalman model is compared with GM(1,1) model and GM-Markov model, and RMSE and MAPE are introduced as evaluation indexes. The results show that the Kalman model has higher prediction accuracy and is more suitable for predicting the lateral convergence deformation of subway shield tunnel.