| This paper combined the The National973Progect"Measurement and ControlTechnology for Segment Erector of Shield machine"(NO.2012CB724306) and Jilin Scienceand Technology Agency’s high-tech project:“Research of Intelligent Integrated PavementDetection Vehicle”, aiming at the key problems of intelligent detection system for pavementdistress, relevant theoretical and experimental researches were conducted, including thedenoising and division of the pavement distress images, extraction and classification of thepavement distress features, real time transmission of the test results etc., and such methodshad been applied to help improve the automation and efficiency of the pavement detectionsystem.The design and structure of the intelligent pavement distress detection system wasdiscussed. The description of the hardware component particularly and the reason forselection of each hardware were also introduced. The installing and hardware debugging, thesoftware of data acquisition system and global positioning system were accomplished.Fucus on the noised pavement image, weak and unclear crack edge information, themultiscale Kalman filter of the pavement image denoising was proposed in the paper. First,the pavement image was decomposed by using wavelet to obtain the high frequencyinformation (includes noise and edges) and low frequency information (includes backgroundinformation) of the image. Second, the high-frequency information was filtered by usingKalman filter for the optimizing wavelet coefficients. And finally, the low-frequencyinformation and high frequency information were reconstructed. As this method combinedthe multiscale features of wavelet transform and Kalman filter estimates in thedecomposition levels, the reconstructed pavement image information was more accurate toachieve a good de-noising effect. Experimental results showed that the multiscale Kalmanfilter method was more effective in de-noising and maintaining the pavement image detailsthan wavelet de-noising method and Kalman filter.Because the edges of pavement distress image were blurred, the fuzzy clusteringmethod for pavement image segmentation was applicable. The improved fuzzy clusteringalgorithm for pavement image segmentation was proposed in the paper. In the algorithm,kernel function was introduced and the OTSU algorithm was used to obtain the optimalinitial cluster centers. This optimized algorithm could effectively avoid converging onlocally optimal and obtain better result of pavement image segmentation. Experimentalresults showed that the improved fuzzy clustering algorithm was more effective in segmentation. There were differences for fractal eigenvalue between background region andcrack region of pavement distress images, and the fractal eigenvalue could be used tosegment the pavement image. And furthermore, the multiscale fractal eigenvalues were moredifferent. Thus, a segmentation algorithm based on multiscale fractal characteristics wasproposed. The variation of fractal dimensions and fractal eigenvalues were used to segmentimages. After experiment, the result showed that the segmentation method based on fractalcharacteristics was effective and feasible. But in the segmented images there were somenoises, and those noises were presented in the form of “isolated” points. A morphologicalmethod was used in eliminating “isolated” points and experimental results showed that theperformance of this method was satisfying.Based on the differences in geometry of different crack types, the features of a varietyof pavement distress were proposed. The features included projection of crack, the regionalstatistical feature and fractal feature of pavement distress, which gave a good description ofthe various types of pavement distress shape characteristic and showed a good performancein image classification. Finally, a classifier was designed based on RBPNN(Radial BasisProbabilistic Neural Network) and four features extracting from pavement crack image asthe input and five type of pavement crack as the output. The lots of image samples as thetraining data were used to train the RBPNN, and then test data were classified by the trainedRBPNN. The result of experiment shows that the classifier gets a better performance torecognize cracks. The methods to compute the length, width and area of cracks wereproposed. The measure reflected the pavement damage at certain degree.In this paper, the availability of introducing wireless communication technology tosupport communication between vehicles in intelligent pavement detection system wasdiscussed. According to the unique features of data transmission in pavement detectionsystem, a packet oriented routing protocol was designed in delay tolerate network and it hadfollowing advantages:(1) the local storage space saving, reduced end to end delay andhigher data success delivery probability, through avoiding the unnecessary datacopying/forwarding and forwarding or neighbor node selection;(2) it could tolerate thecommunication link instability which is required and important in intelligent pavementdetection system due to the higher mobility of the vehicles in this system. The simulationresults showed that, comparing with some previous work (e.g., AODV and Epidemic), theproposed routing protocol could improve the system performance in terms of networkthroughput, end to end delay and data success delivery probability.On the basis of the above theoretical researches and experiments, the database of theintelligent detection system for pavement distress was developed, C++programminglanguage was used to realize the function of the database’s each module. The intelligentpavement detection vehicle was made to realize the collection of the pavement images,distress detection, and remote transmission of the information of the GPS. The engineering application showed that the intelligent detection vehicle could realize real time test, offerclear classification and assessment of the pavement cracks, achieve satellite positioning data,and finish the transmission of pavement distress information. |
PDF Full Text Request