Florian Matusek, Martin Kampel: Shadow Detection based on Chomaticity and Intensity Thresholding

In this work we deal with the problem of shadow detection in image sequences in the context of automated video surveillance. Video surveillance systems monitor an area of interest with the use of video devices. Shadow detection is used in order to avoid problems during the classification of objects. These problems can occur if shadow regions are detected as moving foreground regions and therefore deform the shape of the object. As a result two objects could be "connected" with shadows and are detected as a single object or because of the deformed shape of an object it could be misclassified as another object. After shadow detection the shadow regions are removed and foreground regions with objects are left. Previously developed methods are presented and critically evaluated. Also, test results with image sequences of scenes in different conditions are given.