Improved rail freights cars segmentation method based on line-scan CCD image texture features

We proposed an improved train freights segmentation method based on line-scan CCD freight train images. The captured images were enhanced by the optimized multiple-scale retinex algorithm. Then, the adaptive single-pass seed selection (APSS) algorithm was applied to estimate the initial center values for K-Means segmentation based on Gabor texture features. The different areas of interest were segmented by the K-Means method and finally merged on the basis of the neighbor features. Experimental results show that this method eliminates noises effectively and reduces complexity of computations. In the evaluation of real-time rail freight transportation inspection system, the proposed method receives low error scores of freight segmentation and improves accuracy and efficiency of inspection.