Rolling stock examination is performed by humans to determine the defects during train movement at speeds less than 30KMPH. Computer vision models have the ability to automate this process and this is the goal of this work. Segmentation of parts on a moving train with 4 types of active contour level set models is being performed. They are Chan-Vese (CV), CV based morphological differential gradient (CV-MDG), CV with shape priors (CV-SP) and CV with shape invariance (CV-SI). CV level sets with shape invariance model adjusts the contour according to scale, rotation and location of shape prior object in the rolling stock frame. Visual train rolling stock video data is collected with 240fps high speed sports action camera having wide angle lenses of 52 degrees. Results show level sets produce best segmentation results compared to traditional segmentation methods. The performance indicators of segmented parts form the proposed 4 algorithms are image quality index (IQI) and peak signal-to-noise ratio (PSNR in dB). Total 10 parts were extracted from a bogie using the proposed models and tested against the ground truth models to gauge the performance of the methods. The train has 15 passenger cars with 30 bogies. Further the models are tested under different lighting conditions for 5 different trains. CV shape invariance model produced better segmentations both qualitatively and quantitatively.

train rolling stock examination, image quality index, peak signal-to- noise ratio (PSNR in dB), level set models, shape priors.