| Pavement performance directly affects driving comfort and traffic safety,and it is closely related to surface topography information.Therefore,it is necessary to find a digital pavement topography measurement method with high accuracy and strong operability.Binocular vision technology has become one of the common methods of digital pavement topography measurement because of its simple device,convenient operation and digital storage.However,it is difficult to meet the actual needs of accurate measurement in complex environment due to problems such as the difficulty of accurate matching between pixel points with the same name and insufficient anti-optical interference ability.Therefore,this paper improved the traditional binocular measurement technology,carried out related algorithm research and test effect analysis and evaluation.The main research work carried out in this paper is as follows:(1)Aiming at the low matching accuracy between pixels with the same name of the traditional binocular measurement algorithm,this paper proposes a binocular measurement algorithm under laser constraints.The region to be measured is divided into several sub-regions by laser lines,and forced constraints are formed on the boundary of the sub-regions,so as to enhance the accurate matching between the corresponding sub-regions and improve the overall matching accuracy.The experimental results show that the matching region is reduced by laser line constraint,and the problem of low matching accuracy caused by large global matching region is solved.In the illumination range of 50Lux～350Lux,with the increase of the number of laser line constraints,the measurement accuracy is gradually improved and has good stability.(2)In order to further improve the anti-optical interference ability and the robustness of the algorithm,the Faster-RCNN(Faster-Region based Convolutional Neural Network)object detection model is introduced in this study.The asphalt pavement data set containing laser constraints in the light range of 5Lux～1050Lux was captured and established for the training of laser constraint recognition model.The trained laser constraint recognition model was integrated with the laser constraint binocular measurement algorithm.The experimental results show that the 3D reconstruction effect of the improved algorithm is better than that of the original laser constrained binocular measurement algorithm in the illumination range of5Lux～1050Lux.(3)In order to test the practical application effect of fusion Faster-RCNN laser constrained binocular measurement algorithm,the traditional binocular measurement algorithm,laser constrained binocular measurement algorithm and fusion Faster-RCNN laser constrained binocular measurement algorithm are used to measure four kinds of specimens respectively in indoor environment.The results of random laser point parallax and elevation measurement are compared and analyzed.The measured results show that,under the illumination condition of 5Lux and 1050 Lux,the maximum absolute deviation of parallace measurement is 1.87 pixel and 1.54 pixel,and the maximum absolute deviation of elevation measurement is 0.25 mm and0.23 mm,respectively,for the laser constrained binocular measurement algorithm integrating Faster-RCNN.Are obviously better than the other two measurement algorithms;To further verify the practicability of the algorithm,two areas to be measured were selected in outdoor real pavement environment.Besides comparing the accuracy evaluation of the local laser point point of the three measurement algorithms,Mean Texture Depth(MTD)was also introduced to evaluate the overall measurement accuracy.The results showed that,The maximum absolute deviation of matching parallax and elevation difference in the two pavement areas to be measured are 1.37 pixel and 0.31 mm,respectively.The absolute deviation and relative deviation of MTD measurement values are 0.0508 mm and 8.03%,respectively,which has obvious advantages over the other two measurement algorithms. |
PDF Full Text Request