Research Of Pavement Skid Resistance Estimation Method Based On Texture Features

Road anti-sliding performance has important significance to the development oftraffic safety. The anti-sliding performance of pavement is determined by thepavement texture. And pavement asphalt mixture gradation of macroscopic andmicroscopic structure intuitive performance for pavement texture. So this article isbased on the anti-sliding performance of pavement texture estimation methodresearch.In order to be quicker, more accurate estimates of the pavement skid resistanceperformance, this article will and anti-sliding performance of pavement texture byartificial neural network to establish a connection. To collect pavement texture byusing the texture detection. The platform consists of a laser sensor, STM32F417chipand PC software. After processing data we analyze it for follow-up study. Finallyusing friction pendulum instrument measured set value and the collection of pavementtexture detection platform height value as a parameter of neural network training tobuild the complex non-linear relationship. Then the model estimates pavement skidresistance performance. In this article, the surface texture data are used to dothree-dimensional reconstruction by MATLAB. Measured data for recovering theroad surface texture is not enough. So the interpolation method is used to optimizedata, restoring the pavement texture. At last, that is more ideal.Finally, in this paper, the estimation model was established based on BP neuralnetwork. With the aid of MATLAB auxiliary BP neural network to establish theanti-sliding model, improve efficiency, is better than that of manual calculation. Andalso through the past experience of learning and a new scheme for reasonableinductive to improve its performance. After the construction of the model, theartificial neural network prediction compared with the measured actual value,feasibility test model. After inspection found that the prediction accuracy is not highand slow convergence. So still need to find a parameter. At the same time by usinggenetic algorithm to optimize the initial weights solve the defects of BP network convergence that is slow. After correcting model, neural network prediction errorcontrol in less than5%, and the convergence speed is doubled.