رامين فتحي

In this thesis, the topic of obstacle detection on the train tracks is discussed using image processing and deep learning. The purpose of this research is to provide an efficient and accurate method to identify and locate obstacles that may cause danger and damage to trains. For this purpose, a combination of convolutional neural networks and transfer learning networks has been used. The proposed method includes two stages of zoning and obstacle detection. In the segmentation stage, U-Net, SegNet and YOLOv8-Segmantation net- works have been used to extract the rail area and needle direction. In the obstacle detection stage, YOLOv8, SSD and Fast_RCNN networks have been used to identify and locate various obstacles such as humans, animals, cars, rocks, etc. The results of the eva‎luation of the proposed method on a database of railway images prepared with different environmental conditions show that the proposed method has high accuracy and speed. And as a result, it can be used as a suitable solution to detect the obstacle on the train track. In this work, network training was done in two stages. In the first stage, it was done without considering the environmental conditions and in the second stage by applying the effect of environmental conditions on the data. In the first stage, when the network is not familiar with the data of different conditions, the accuracy in viewing these images increased to 81.9%, but when it was trained in different environmental conditions, the accuracy increased to 91.4%.

تشخيص مانع , يادگيري ماشين , پردازش تصوير , ريل قطار

obsticle detection , machine learning , image processing , train rail track

ramin fathi

ahmad mirabadi