Research On Typical Train Fault Image Recognition Method Based On Shape Descriptor

Trouble of moving Freight car Detection System is a set of independent research and development in our country’s dynamic image detection system, which based on machine vision, it aims at eliminating the artificial detection uncertainty factors, improving the inspection accuracy, and advancing inspection way to automatic detection. The missing fault of side frame key and truncated plug door handle closed two kinds of common faults are discussed and researched in this paper. According to the characteristics of single color and complex background of the TFDS image, using their shape characteristics, a typical fault recognition scheme is presented based on the shape descriptor.In the image recognition algorithm for side frame key of train based on shape context, a regional image with side frame key is taken as a template in searching in the TFDS images tested. The shape characteristics of these regional images are described by shape context. A new shape distance is defined as the similarity index for image matching by weighting shape context distance and bending energy. Accordingly, the judgment of missing side frame key is made if the similar region in the TFDS image is searched by the template. Through a large number of testing images experiments using Matlab programming, the testing images with side frame key can be judged well by choosing an appropriate threshold as the shape distance threshold.In the image recognition algorithm for truncated plug door handle closed of train, respectively choose distance function and improved distance function as shape descriptor, shape matching using aphelion as starting point and the dynamic match two ways, four kinds of handle closed fault recognition algorithm are acquired. The four kinds of the implementation process of the algorithm are introduced by the aphelion matching algorithm, which based on improved distance function as example. The handle contour is taken as a template, in order to outline for object matching, according to the perimeter of geometrical characteristics of the contour to simplify the test image, contours are described by improved distance function and the characteristic vector is acquired. The shape of the distance between contours are calculated by using the Euclidean distance as the similarity index for judging truncated plug door handle closed and the above four kinds of algorithms are realized by using VC programming. The results of the experiments show that the algorithm based on the improved distance function has better degree of differentiation and adopts aphelion as a matching starting point for the execution time is shorter, so the aphelion matching algorithm based on the improved distance function is more suitable for handle closed fault identification. Final template matching based on gray level is introduced, and compares with aphelion matching algorithm based on the improved distance function, the latter is more efficient and has stronger robustness and stability.The missing fault of side frame key and truncated plug door handle closed two kinds of common faults as example in this paper. The effectiveness of the shape context and the improved distance function of two shape descriptor in typical fault TFDS of image recognition are verified.