Machine vision (object recognition) is the ability for recognizing images and to understand what is seen. It involves digital cameras, digital signal processing and amachine learning algorithm. After the image is taken the particular steps within machine vision include:
Image processing - stitching, filtering, pixel counting.
Segmentation - partitioning the image into multiple segments to simplify and/orchange the representation of the image into something that is meaningful and easierto analyze.
Blob checking - checking the image for discrete spots of connected pixels (e.g. a blackarea in a grey object) as image landmarks. These blobs frequently represent opticaltargets for observation, robotic capture, or manufacturing failure.
Pattern recognition algorithm including template matching, i.e. finding and matchingwith specific patterns using some ML method (neural network, deep learning etc.). Re-positioning of the object may be required, or varying in size.
Railway Use Cases
To maintain the railway, it is important to automatically inspect different assets andcomponents, such as connections among railway tracks, and contour poles supportingpower cables. Tracks can be damaged by the friction between their surfaces andwheels. Electrification systems, such as overhead power lines, should be periodicallychecked since they should stably supply electric power to the locomotive.
However, traditional inspection systems are always dependent on human manualobservation, which is not quite efficient, since human observers become easily tiredUsing Images to Identify Things and Assets in Railway systemsPage 2 of 2and lose focus on important objects after a few minutes. Also, only a small timeinterval is allowed for human inspection, since trains usually operate almost all day.
To solve those problems, various image processing and computer vision-basedautomatic inspection approaches have been introduced.
The Machine Vision Approaches
The new AI/ ML-based techniques, such as computer vision and image recognition canbe applied to inspect defects in railways for safety and maintenance, and it is calledimage-based railway inspection system (IRIS). Here we will list some significant applications of machine vision (object recognition) in railway transport:
In 2017, Amaral and his colleagues presented a system for obstacle detection inrailway level crossings - by a set of points obtained from curved 2D laser scanners.
Kim and Cohn (2004) set up a camera in front of a locomotive to investigate the levelcrossing traffic accidents. They developed a computer vision system that automatically detects the possible after-accident scenes by detecting the shape of the vehiclespassing in front of the train.
Kantor and colleagues, for maintenance purposes, applied a laser light line to generatea 3-D profile of the railroad surface, and a ground penetrating radar to obtainsubsurface measurements.
For visual inspection - Rubinsztejn and Chen used cameras to acquire real images. In order to achieve the automatic detection of parts of interest, missing elements ordefects, they processed the captured images with pattern recognition algorithms.
Singh et al. (2006) used image processing methods, such as edge detection and coloranalysis, to detect missing clips.
Deutschl et al. (2004) used convolution filters and morphological image analysis todetect rail surface defects.
Weil combined a ground penetrating radar with infrared imaging systems to detectsubsurface defects in railroad track beds. The hardware and software for machine vision have improved dramatically in recentyears, so its applications in railway technology will increase in the upcoming years.
