Automatic Railway Track Element Detection for East-West Route of Addis Ababa LRT

In railways, regular maintenance and inspection of railway track is quiet commonly used approach to maintain an acceptable level of safety. Such approach is currently done in most railway companies; ERC will not be a different, using a very laborious, lengthy and manual work by visually inspecting and visiting test sites with a help of trained operators. Another approach is to have an automatic train control system where a train vehicle is completely controlled by a central control system without the need of human driver intervention since computer-aided machines are less likely to make mistakes than humans once carefully designed and installed. In this thesis, an automated system is presented which mitigates the said problem by detecting a railway track elements (such as fasteners) from a real railway track image taken by a digital camera that is mounted in the train. Based on the level of difficulty in the detection problem, suitable combinations of image processing and pattern recognition methods are applied to achieve high performance automated detection. The system is primarily based on the Viola-Jones object detection framework Viola and Jones, but it selects best new Haar wavelet-like features more convenient to the problem of Addis Ababa Light rail transit track fasteners detection. The performance of the system is improved by employing an appropriate image pre-processing and performance evaluation techniques. Accordingly, the system has managed to achieve a 94.6% detection rate of performance and has a failure rate of 0.11%.