Esther Resendiz


Esther Resendiz is a Ph.D. candidate in the Department of Electrical and Computer Engineering at the University of Illinois at Urbana-Champaign. She is a member of the Computer Vision and Robotics Laboratory (CVRL) in Beckman Institute. She primarily focuses on signal processing-based methods for image and video analysis. She collaborates frequently with the Railroad Engineering Program in the Civil & Environmental Engineering Department, and is currently applying machine vision to track inspection. Her previous work includes undercarriage inspection for passenger railcars. She is also the President and co-founder of Fashion Latte Inc., a visual search startup for the online apparel domain that she founded in 2008. Prior to joining the CVRL, she received an M.S. in Electrical and Computer Engineering from UIUC and a B.S. in Electrical Engineering from the University of Texas at Austin.


Railroad practice and FRA regulations require regular, visual inspection of track in order to monitor its condition and detect potential defects. Machine vision has proven valuable in a variety of railroad inspection applications due to its ability to quickly, accurately and objectively record and process large amounts of video and image data. Over the past decade, machine vision has begun to be applied to inspection of railroad track components with inspection of certain parts of the track system already implemented. However, other aspects are more challenging and are still in various stages of research and development. Among the latter is the condition and position of cut spikes and rail anchors. To detect these components it is useful to identify larger components such as ties and tie plates and use these as reference points to identify other components and characteristics. Many of these components occur periodically and signal-processing techniques can be used to detect and segment them. To accomplish this, an image or video is converted into one-dimensional signals and then spectral estimation is applied to those signals in order to detect periodicity. The periodically repeating objects are then detected and segmented.

This seminar will present research on development of an unsupervised method for detecting periodically occurring components that are repeating in one direction. In railroad track inspection, detecting and extracting the periodic components that appear in inspection images is useful because these components are often indicative of overall track condition. The method will be demonstrated on track inspection panoramas and on turnout inspection videos. Then, a more general method will be described that detects and localizes periodically occurring objects in images that repeat in arbitrary directions. In this method, it is assumed that an object repeats along one unknown direction within the two-dimensional image. The goal is to detect that direction of periodicity, and to localize those periodic objects within the image. This method will be illustrated on example images, including track inspection panoramas.

Machine Vision for Railroad Track Inspection