IMAGE ANALYSIS FOR DETERMINING THE PROFILE OF RAILWAY TRACK SURFACE USING LINE LASER

• Main Author: PRIMADI (NIM. 23814002), ARDIAN
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/25723
• Institution: Institut Teknologi Bandung

Indonesian Government has been targeting to accelerate the construction of railway track. If in 2014 there are 4.969 km of rail tracks are active then targeted in 2030 rail tracks are active to be along 12.100 Km spread all across Indonesia. This target represents government’s intention of making train as one of the key transportation mode in Indonesia. One of the main concerns is the safety enhancement for railway transport modes. Based on data from the Bureau of Communications and General Information of the Directorate General of Railways, train collapsing accidents due to track damage is the most frequent case of train accidents which have 113 (71.51%) cases of 158 cases of train accidents occurring between 2010 and 2014. One of causing factor of the train collaps is track inspection system that has not been optimized. <br /> <br /> A conventional method is still implemented in inspecting track surface that is by officers walking along tracks and using their senses of vision to measure the quality of the track. The method is very tiring, time consuming and allowinga lot of measurement error. Therefore, it is necessary to develop a technology by using computation system and image processing technology to obtain the objective. One which has been developed is track inspection system by using laser vision. This system is performed by determining the profile of studied track, so that each profile alteration can indicate the condition of the track. In previous research, the image acquisition process was obtained by configuring digital camera sensors, track and laser lines in triangulation configuration and save the result in video data. This research determines the profile of track surface by analyzing the image of line laser beam on recorded in the video. <br /> <br /> Firstly, determination of the ROI (Region of interest) is conducted to choose the focus to be analyzed from the digital image. Vibration on the device in measurement process causing the shift of camera’s field of view, hence the position of laser beam image area on each digital image series in the video is also shifted. In previous study, the determination of ROI was done statically by selecting certain area based on the changing area of laser beam image position. Consequently the selection of static ROI is larger than needed. Dynamic ROI selection by using template matching method done in this research is purposed to determine the area of laser beam of the digital image automatically for each frame. Static ROI method obtain 7,89% area of a frame, meanwhile dynamic ROI result in only 1,38% area of a frame. From the analysis of both Static ROI image and Dynamic ROI image from 50 ROI data, it is obtained that the average comparison between the dimension of line laser image and the dimension of ROI is 6% and 33%. From image segmentation on both ROI with 200 data, the image that can be well segmented is 44.5% on static ROI and 98.4% on the dynamic ROI. <br /> <br /> Image processing is applied to determine the main framework of line laser on the track surface. In previous research, the image processing is applied with three steps of operation, i.e. convolution, thresholding (Otsu) and skeletonizing. However, light intensity of the environment contaminates the intensity of laser beam, such that the intensity of laser beam which is too dim disappears on thresholding step. In this research, the image processing is applied by the following operation steps: convolution, maximum intensity determination and interpolation. The length of the main framework profile of laser beam is devoted as a parameter to represent the characteristic of track surface profile quantitatively. Comparative analysis technique of image processing is done by comparing the parameter value resulted from the previous research with the value from the current research. On the 600 analyzed digital data, it is obtained that the average length of the profile of laser beam main framework from previous-research method is 253 with average deviation of 11 pixels and standard deviation of 6.34%. Meanwhile the result on this research is 271 pixels, with average deviation of 9 pixels and standard deviation 5.50%.