Optimal Threshold Image Segmentation Method Based on Genetic Algorithm in Wheel Set Online Measurement

Online measurement of wheel set wear parameters is important for train safety. The acquisition and processing of wheel set profile image is a key problem in online measuring based on machine vision. Appropriate threshold segmentation is needed to extract clear wheel set profile curve from varying background. Otsu algorithm is the traditional optimal threshold method, which is popular and efficient, but when target is much smaller than background, or SNR is low, the segmentation effect is poor. An optimal threshold segmentation method based on entropy criterion and genetic algorithm (GAEC) was proposed according to analysis of the influencing factors in image acquisition. The factors included environmental illumination, light source life, light reflection, CCD exposure time and image histogram characteristics. Experiment results showed that the proposed algorithm can effectively eliminated interferences in the acquired images. Searching efficiency of the optimal threshold and the segmentation quality were also improved.