Research On Pavement Crack Detection And Recognition Methods Based On Image Analysis

Pavement crack detection is an important part of the pavement quality management. With the continuous development of automation level, the pavement crack detection and recognition method based on image analysis has become development trend of pavement detection. The crack detection is more difficult as the pavement crack is easy to be disturbed by the noise and various road conditions.Based on the analysis and summary of the research of some scholars and from the practical point of view, the key technologies of pavement crack detection and recognition were studied and the main research contents are as follows:Pavement crack image enhancement technologies were researched, improved median filter united with wavelet threshold filtering method of were applied for crack image denoising. Especially, for pavement crack image affected by uneven illumination, a comparative study of the homomorphic filtering method and uneven illumination restraining approach based on wavelet transform was conducted.Crack object extraction methods were studied, traditional image segmentation methods, such as edge detection operator, threshold segmentation algorithm were applied to investigate the application effect in the crack image segmentation; multi structure element anti noise morphological edge detection operator was used for pavement cracks target extraction. And the method of combining OTSU, local gray feature, dissimilarity feature and orientation feature extractions with multi-structure element morphology operator for denoising was presented and the experiment results show the good performance of object extraction and better stability, and the requirements of road pavement crack detection were met.Four types of pavement cracks were analyzed, and feature vectors based on the projection transform and mathematical statistics, the breakage distribution density of the crack were extracted and the extraction method of the distribution density feature was improved. Three crack classifier were designed based on the BP neural network.Experiment results show improved distribution density feature classifier recognition performance is better than the other two classifiers, but the nonlinear crack classification recognition rate still needs further improvement. According to the three classifiers’ performance, a fused classifier was designed, and the experiments show the advantage of the classifier.