Abstract
Abstract. Anomaly detection in imagery has widely been studied and enhanced towards the requirements of today’s available sensor data, whereas many of them require a background estimation in order to identify an anomaly or target. In this paper, we examine an analysis of simulation as background estimator for anomaly detection in thermal images of urban sceneries. We generate a surface temperature image and a sensor-like infrared image by combined image and elevation data and a thermal model suited for large scenes and fast simulation. With the simulated thermal image, we define anomalies as deviation between measurement and simulation. Pixel-wise image differencing of the measured and simulated temperatures and infrared images respectively are performed and evaluated concerning the full images as well as class-wise, including a material classification of the observed area. Our approach shows complementary results compared to RXD application on the measured infrared images. Metal roofs which appear warm in the thermal image and are not visually distinguishable from the residual image are detected.
Highlights
AND PREVIOUS WORKAnomaly detection algorithms have widely been studied and enhanced towards the requirements of today’s available sensor data
The field of application thereof spans from civilian to military. Whilst the latter may deal with target detection in defense systems, e.g. small targets sensed by a shipborne warning system (Hui et al, 2016), or vessel detection (Islam et al, 2009), civilian applications include fire detection (Guo et al, 2016, Kato, Nakamura, 2017), earthquake analysis (Wu Lixin et al, 2006), urban heat islands (Sharma, Joshi, 2014), and many more
Our study focuses on the basic usage of simulation for anomaly detection, and outlines the similarities and differences to conventional anomaly detection by application of the RXD on long-wave infrared (LWIR) images
Summary
Detection algorithms have widely been studied and enhanced towards the requirements of today’s available sensor data. Many of them base upon machine learning, resemble classification approaches and try to define a normal behaviour in order to find data points deviating from it. When it comes to anomaly detection in images visualizing sensor data from other spectral ranges than the visible, statistics-based techniques appear to be the method of choice (Guo et al, 2016). The local-RXD (Gorelnik et al, 2010), for example, network approaches require a preceeding target labeling in the training datasets, which makes them difficult employ if appropriate training data is not available or if a target can not be defined
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callMore From: The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
