Research On The Fastener Status Recognition Based On LDA Combined With Structure Information

As the key connecting accessories for rails and sleepers, railway fastener plays the crucial role in railway safety. However, the fasteners may lose or break because of the vibration caused by rushing trains and other factors, so fastener status detection greatly concerns the railway safety. Huge labors and material resources will be wasted if we totally adopt the traditional detection methods, which will hinder the railway development. Detection method based on machine vision shows much more effective and has become the main international detection method.Fastener detection based on computer vision technology has been researched. The features of fasteners images have been extracted through the digital images collection and image process algorithms. An improved LBP combined LDA method was proposed to achieve the fastener status recognition. The main contents are as follows:1. The present situation of railway detection was stated and the fastener status recognition algorithm based on computer vision is simply introduced. According to fastener images features, image preprocessing methods is researched to highlight the fastener features.2. Considering original LBP is unable to highlight the edge features effectively, an improved LBP coding algorithm for fastener images is proposed. The proposed algorithm utilize the thought of the second order derivative can highlight the image border features. Firstly, the mean gradient of local neighbor pixels and the differences between local neighbor pixel and center pixel are calculated; then the original center pixel is replaced by the above mean gradient; finally the above differences are compared with the new center pixel to get the improved LBP codes. Experiment results show that the improved LBP codes can clearly reflect the gradient changes and show the image features. The coded images are more easily for fasteners classification and the detection accuracy is improved.3. To solve the problem of the lacking fastener position information after LBP coding, the position information is remained through partitioning the coded image. Meanwhile, through second partitioning, the corresponding position information of words bag is added; the words bag construction of LDA is improved and the LAD description for LBP coding image is realized. Finally the fastener status is recognized with SVM. Experiments show that the improved method can recognize the fastener status more accurately.