Lu Zhaijun

Correlation matching algorithm based on dynamic gray threshold for catadioptric stereo vision

A catadioptric stereo vision system composed of a CCD sensor, a lens and two hexahedral prisms was developed for detecting the offset of railway vehicle relative to the track. In this way, two images of the same track can be gotten from the imaging plane of one camera. In order to obtain the parallax necessary for recovering the depth information from every frame image, the optimal evaluation function of the dynamic gray-scale threshold was constructed and the maximum correlation matching algorithm was employed. Combining some features of images, a gray threshold can be obtained through the function. A matching template can be intercepted from the gray curve of every frame image. By calculating correlation coefficients between the matching template and the whole gray curve, the authors get a correlation coefficient curve. The distance between the two points whose correlation coefficient respectively is the maximal and the second maximum is the parallax. The experimental result shows that the maximal correlation algorithm based the optimal dynamic gray threshold is robust and can be adaptive to environmental change.

Measuring Technology of Rail Vehicle's Offset Caused by Vibration

On-line measuring the displacement of the running rolling stock relative to the track, is the base that rationally determines leaving space between the train and the structure limit. The scheme based on machine vision for measuring the displacement is brought forward. Four dynamic offset sensors are non-collinearly mounted on the car-body. Each sensor composed of a linear Charge Couple Device (that is the CCD), a lens and four mirrors is a monocular stereo vision system. In this way, two images of the same track can be gotten from the imaging plane of one camera. With the movement of car-body, the cameras scan the tracks, and images are analyzed real-timely in order to get the lateral and vertical displacements of those cameras relative to rail. Considering that we can only obtain two dimensions data through binocular linear CCD stereo vision, the reasonable locations those four sensors are mounted on are important. According to installment requirement of cameras put in this paper, the attitude of the car-body can be obtained by analyzing displacement of those locations and the vertical and lateral displacement of random location will be got through further calculation. Detection principle is introduced, and some formulas those calculate five variables describing car-body attitude and the displacement of random location are built.