Reasearch Of Key Components Detection Algorithm Of Locomotive Running-gear Based On Implocit Shape Model

Since the 21 st century, with the rapid development of China's economy, a rapid development also happened in railway transportation which is one of the most important transportation method in China. With the development of railway transportation, a lot of attentions also paid to its safety issues. Running-gear is a key part in train, its operation state is closely related to transportation safety, and therefore, anomaly detection of running-gear is one of the most important parts in the daily maintenance. In the early days, the maintenance of running-gear was mainly artificial, which is not only worked hard but also occupied the running time of train. Later, with the development of machine vision field, an imaging system was produced to capture the images of running-gear, and the maintainers can inspect it by watching images.It is improving the working efficiency by avoids the workload of getting under the car. But,it is unreliable for detection by artificial due the visual fatigue. So, an automatic abnormal inspection system is very urgently for running-gear.The determining of anomaly is the most important and challenging work in this imaging system. And the recognition and localization of key components is the first step of this work.Therefore, we analyze the difficulty of object localization in crowded scene, and a top to bottom object localization based on "prediction theory" was proposed. The proposed method can be used to solve the problem of rapid object detection in structured scene. This article mainly works as follows:1. For the representation of local images, sparse representation is one of the best method as now on. The phenomenon of "dead neurons" happened frequently, and the cluster result is extremely easy to get to the local minimum in the traditional sparse representation method. Because there is no control of the effectiveness of every atom.For this kind of situation,a clustering method based on the competition mechanism was presented in this paper. In this method, the efficient of atoms was guaranteed by control the number of samples used to train every atom.2. The second step of image representation is using the atoms trained to represent all the local images. The traditional method is calculating the distance between local feature and atoms, which is simple and straightforward. However, due to the high degree of similarity between the basic image blocks, the final classification result is not stable.Therefore, "inhibition of similarity between classes" method was introduced in this paper. By inhibiting the similar components method to achieve outstanding differences between classes,enhance the classification of stability and reliability.3. It is easy to capture a lot of disjointed and inefficient edges while edge detection of running-gear image, due the effected by light,waterlogging,and dirt. Therefore,based on the theory of visual system,an improved edge detection algorithm was presented in this paper. In this method,the effective edge characters are trained by USF edge sets,and the effective edge characters can be used to evaluate the effectiveness of edges detected. Then, the invalid edges can be eliminated and the reliability of edges can be improved. The experiment shows that the proposed method can be used to wipe out the effecting of noises while edge detection.4. In the abnormal detection system of locomotives, the identification of key components is one of the most important work. Based on the implicit form generation model and prediction theory, a top to down object recognize and localize method was presented in this paper. The method can be used to solve the problem of fast and precise target identification for structurally stable targets.5. At last, an anomaly inspection system was presented in this paper. Based on the object recognize and localize method presented in this paper, it can be used to detect the anomaly happened on the key components. The experiment shows that, the accuracy rate of our method has a significant improvement compared with traditional method,has broad application prospects.