Identification Of Power Locomotive Maintenance Instrument Based On Machine Vision

Identifying instrument is an important technique for realizing the electric locomotive maintenance intelligently and digitally. Intelligent identifying instrument based on machine vision should be with roboticized capacities: instrument classification; recognition the character of instrument; put measurement data in storage. However there are some shortcomings: without instrument classification; can’t catch the instrument’s character underlying complicated illumination conditions; didn’t fulfill the recognition character with geometric deformation. Thinks to these insufficient, the interested points for our research including instrument classification, achieve and recognize the instrument’s character underlying the condition of changing. The detailed researching content consists of three points:(1)The local threshold segmentation with adaptive window width and strategy of evaluating binary imageUneven illumination result in text losing and false text, with a view to this, we add the adaptive window width to adapt to Niblack algorithm, at the same time, add the strategy of evaluating binary image to define the terminal condition for image binarization, optimizing text mark by these. The modified method of local threshold segmentation is good to catch text and delete the false text.(2)Text recognition based on Zernike momentsBecause of the text come from work environment are rotated, translation, different size and deformation, we issue take the Zernike moments to describe text, due to the modified Zernike moments is invariant for rotation, translation, size and deformation. After standardizing Zernike moments feature, we employ the nearest-neighbor classification to recognize text. Under the circumstance for text is achieved perfectly, the experimental result show the accuracy of text recognition satisfy the requirement of technical protocol.(3)Instrument classification based on color constancy and Zernike moment of textInstruments with rich color information and so many styles for instruments display. Based on these two characteristics, we define a color constancy feature because of Gaussian’s color model, in addition, we take color constancy feature and Zernike moments feature of text to combine into a feature vector. This vector is used to instruments classification. Test result show a good achievement underlying our instruments library.