An improved detection method for railway fasteners

Aiming at the disadvantages of low efficiency and poor stability of the existing methods of fastener detection, this paper proposes a method of fastener detection based on template matching. Firstly, the image is processed by noise reduction, this method does not need to be based on histogram can match template directly. We choose a standard fastener image as template, read template firstly, then slide the image blocks on the input image to match the template and the input image. Experimental results show that this method can effectively detect the flaw, deformation and obstacles of railway fastener occlusion. This method has a fast matching speed and a good robustness, it’s detection accuracy is up to 96%


Introduction
Railway fastener is used to tightly fixed tracks to the sleepers, as an important part of the railway system, plays an important role in ensuring the high-speed running of the railway train.At present, with many kinds of literature are used as the detection method of fastener.Ça˘glar Aytekin, Yousef Rezaeitabar et al [1 ] using a method which fuse various methods based on pixel-wise and histogram similarities.The method has a large amount of calculation and low accuracy.Xavier Gibert, Vishal M. Patel et al [2] proposed a method of railway fastener state detection which is based on the HOG feature and PCA algorithm.This method need to calculate the histogram of each image, a large amount of calculation, and a large number of training.The efficiency of this method is not high.Liu Manhua, Yang Jinfeng et al [3] take use of directional field algorithm to extract parts of the image features, and then do the template matching.This method has higher computing speed, but higher false detection rate and low accuracy.Stella, Mazzeo et al [4]and Marino Distante et al [5] using Wavelet analysis and, Multilayer Perceptron Neural classifiers to detect hexagon Al-headed bolts (a kind of fastener reveals the presence/absence), of the fastener P -296 bolts with high detection and classification rate, but that is not applicable to our situation.Because railway fastener has a complex background.Characteristics of fastener has a obvious difference from characteristics of track, subgrade and sleepers.distinct characteristics.Considering the shortage of existing methods, this paper proposes a new method for fastener detection.Using image and standard template matching method.Calculate the similarity between the template and the template.When the threshold value is beyond a given threshold, the fastener can be identified as missing, deformed or blocked.

Fsterner detection system design
Fastener detection system is mainly composed of the bottom fastener image acquisition device and the software processing system.Fastener image acquisition device is mainly responsible for accurate, complete, clear to capture the image of each fastener.Specific devices such as figure 1:

Fig 1. Fastener image collecting device
LED light source to provide the system with sufficient brightness and uniform light, reducing the impact of light and weather on the picture quality.A high speed industrial camera with a high frequency can be guaranteed to capture a clear image even in the case of a vibration.Photoelectric sensor is used as the trigger sensor of the fastener.When the train through each fastener, the sensor input a signal to the control system, the control system output a signal to the camera, shooting a pair of images.In this way, we can ensure that each image can only be collected to a complete image of the fastener.To avoid the image caused by the false detection and miss detection conditions.Main control board and circuit inside the box body.

Fastener detection algorithm
The method for detecting the fastener is as follows: select an image of the complete fastener on the actual line as a template, and the template is sliding from the upper left corner to the lower right corner of the image.Calculate the similarity of each location.Match The best match position is the fastener position.If the acquisition image size is W*H and the template size is w*h, the output image size is (W-w+1) * (H-h+1).The coordinates of a pixel point on the image acquisition image is (x, y), and the coordinate value of a pixel point on the template image is (x ', y').T (x ', y') represents the position of the template (x ', y') of the pixel value.I (x ', y') representation (x ', y') to cover the pixel value of the corresponding point on the image.R (x,y) represents the final calculation results.The following algorithms are given in this paper:

SQDIFF NORMED:
This kind of method uses the square difference to match, the best match is 0 but if the match is bad, the match value is bigger.

CCORR NORMED:
This kind of method uses the multiplication operation between the template and the image.Therefore, the larger number of representative matching degree is higher.0 represents the worst matching effect.

CCOEFF NORMED:
In the algorithm mentioned above

P -297
This method can match the relative value of the template to the average value of the template and the image to the mean value of the template.1 epresents the perfect match result, and the -1 represents a bad match, and the 0 represents no correlation.

Experiments
Please provide a shortened running head (not more than four words, each starting with a Capital) for the title of your paper.This will appear with page numbers on the top right-hand side of your paper on odd pages.Using the above three algorithms, we use 77*137 pixels of the fastener image template, 300*400 image matching, the matching effect as shown in figure: Using the above three algorithms, when the output value is close to the ideal result, we judge for the fastener is normal, when the output result is abnormal, we judge the fault or obstructions to fastener.Due to the small number of samples, we use the method of artificial judgment to judge.In the follow-up work, we will use the PCA algorithm to classify and identify, in order to achieve the most effective.This experiment for 200 pictures, including normal picture 177, abnormal picture 23, including the fastener deformation 8, lost 4, the barrier 11, the statistics of the three algorithms matching results are as follows: CCOEFF NORMED 99% 0 1%

Summary
By using these three methods, template matching effect can be achieved.From the SQDIFF NORMED matching method to the CCORR NORMED matching method, and then to the CCOEFF NORMED matching method, the matching accuracy is improved, but the corresponding calculation time is also increased gradually.Due to the large size of the fasteners detecting data, we require the higher accuracy of matching result.The NORMED CCOEFF algorithm is more suitable for the detection of fasteners.We will further optimize the efficiency of the algorithm in the following work, select the high performance computer configuration, in order to improve the computational efficiency. .