Analysis Of Image Recognition Algorithm For The Freight Car’s Failure Based On Its Salient Feature

TFDS is a popular security assurance system that is demanded by the department ofrailway now. Both the efficiency and the quality of this system are superior to those oftraditional static detection, but its currently working model should change from computercombining with human-control model to the final realization of the completecomputer-control model. This paper is intended to design and implement the automaticrecognition algorithm for six failures by using image processing techniques and computervision theory, to push forward the transformation in the working model of TFDS.This paper only discusses the processing method of the topography of every failure.Based on the thoery of salient features, six failures are able to be divided into three groups:the significant change in region、the significant change in boundary and the significantchange in structure. Accordingly, the main contents of this article are as follows:As for failures belonged to the significant change in region, the retaining bound lost、the oil throw, recognition algorithms of them have been analyzed according to their ownregion features. For the segmentation of the retaining bound, a method of thresholdestimation based on the priori knowledge has been proposed, it is capable of overcomingunder-segmentation problem resulted from nonuniform illumination while using theOTSU. In addition, characteristics extracted from the envelope of the projection histogramand the approach of grey level value statistic based on local window are used respectivelyto describe the shape feature of the retaining bound and the gray distribution feature of theoil throw.As for failures belonged to the significant change in boundary, the binding boltloose、the slave plate warp、the draft sill warp, their boundary features have been seriouslystudied to improve recognition algorithms. For example, a corner detection method basedon the gray-scale projection has been presented to extract the boundary of the slave plate,it performs well for the image with weak gradient feature. Moreover, an edge followingalgorithm based on the fan-shaped search has been designed in this paper at the aim of theboundary extraction of the draft sill, which is able to follow the curve and connect the gapbetween discontinuous edges at the same time. As for the failure belonged to the significant change in structure, the coupler yokekey lost, the symmetrical structure of it helps a lot for the recognition algorithm. A meanof structure location based on template matching has been put forward to segment the ROIin a short time.Overall, the mentality and the implementation procedure of recognition algorithmsfor six failures have been elaborated one by one, and then algorithms have been proved oftheir advantages in the efficiency and real-time performance through testing of largenumber of image samples, meanwhile, the rate of wrong detection and the rate of missingdetection can both achieve the expected aim of automatic recognition for the failure offreight cars.