Deep Learning and Multi-Model Fusion for Analyzing Staircase Wear and Human Traffic Patterns

This paper aims at providing scientific guideline for archaeological research by analyzing the relationship between stair wear and human activity patterns. The Archard Model is adopted to build staircase wear model to take account of human activities and natural influences. A Dynamic Wear Differential Equation is derived and the Numerical stability is ensured by spatial and temporal discretization, explicit Eulerian and CFL conditions. The maximum, minimum and average wear depth are 1.5568 mm, 0.0668 mm and 0.3452 mm, respectively, which is consistent with the assumptions of human activity habits and Gaussian distribution of pedestrian flow. A Multi-Source Information Fusion Model, which employs a simplified Archard wear equation, is constructed to predict human activity patterns. A Gradient Boosted Decision Tree (GBDT) was applied to fit the model, and 5-fold cross-validation and Bayesian optimization were used to improve the prediction accuracy. The usage frequency of the stair can be estimated as 493 persons/day. To predict the walk direction and pattern, the longitudinal and lateral features are defined, and the Gaussian Mixture Model (GMM) and Bayesian Classification Framework are applied. The obtained results show that the peak spacing: 0.39 m and the asymmetry index: 0.10. A Multi-Angle Analysis of Measurement Model is proposed to optimize the measurement data collection. The sampling density function, error evaluation model and time measurement model are established and solved iteratively. The results exhibit high accuracy with average measurement error of 0.003 mm and total measurement time of 421.55 hours, which indicates the data obtained by the proposed measurement scheme is reliable for the subsequent analysis. To realize the time series reconstruction of the staircase by the Markov Chain Monte Carlo (MCMC) method, the Bayesian Temporal Inference Model and Multi-information Historical Reconstruction Model are developed. The factors which have great influence on the staircase wear are considered including the intensity of use, environmental impacts, and maintenance activities. To validate the proposed model and method, and the given dataset is adopted. The age of the staircase is estimated to be about 219.2 years, with an error no more than 40 years. The model was also effective in analyzing the reliability of the renovation, material sources, and historical records with a reliability index of 0.8. The sensitivity analysis is performed to quantify the prediction error using Mean Absolute Error (MAE). Compared with the impact of human activity intensity and path distribution standard deviation, the diffusion coefficient has a greater impact on the accuracy of predicting the frequency of stair use.