Abstract
The subject of the work is in the field of creating autopilot railway transport. One of the problems related to railway safety is considered. This paper presents a developed algorithm for determining the railway track from a video image. It is the first step in determining the free path or obstacles in front of the locomotive. The proposed algorithm is a combination of several machine learning algorithms that are applied sequentially (boosting). The first stage of the algorithm is the extraction and classification of features from the image. In this stage, the speeded up robust features or SURF-method is used. At the output of the SURF-stage, we obtain information in the form coordinates of key points and their descriptors. In the second stage, the selected key points are classified. Combinations of two classification methods are used: the K-nearest neighbors or KNN-method and the support vector machines or SVM-method. The final step is the compilation of a railway track mask. For this, the nearest neighbor graph method is used. For practical use of the found mask, the inverse perspective transformation is performed. The efficiency of the developed algorithm is shown experimentally. It can be considered as one of the ways of image segmentation. The main advantages of the algorithm are associated with minimal preparation of the training sample and the ability to analyze its work for further improvement. The results of processing real video images obtained from a video camera mounted on a locomotive are presented.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have