Context-based region labeling for event detection in surveillance video

Context-based object-of-interest detection for a generic traffic surveillance analysis system

“…The region labeling stage explores the local features (color and texture) as well as global features [7]. In the frame- work, spatial location in the image is considered as a global feature, which reduces the ambiguities in contextual information from smooth regions, such as sky and water.…”

“…For each class, we measure the percentage of pixels classified as belonging to this class, in a given region. We then assign a specific label to a region when its percentage of positively classified pixels is above a threshold [7]. Considering the scenario of traffic surveillance, we train 5 classifiers: sky, vegetation, construction/buildings, road and water.…”