Moving objects detection and labeling in surveillance videos
Metadata[+] Show full item record
Moving objects detection and labeling are an important component of intelligent surveillance systems. Accurate classification of moving objects allows the monitoring system to react to certain events of interest. In this work, we propose a reliable system for moving objects detection and labeling in traffic surveillance videos. We use background subtraction to obtain the foreground of moving objects and the positions of objects are determined after connected components labeling. To separate the connected moving objects, we propose a novel method by combining labeling shape context and color distance between two separate parts of the original image. The experimental results show our methods achieves good accuracy. To classify object types, we use a combination of HOG (Histogram of Oriented Gradients) and LBP (Local Binary Pattern) with BoW (Bag of Words) models as the feature representation. In the part of objects color classification, we combine color features extracted from RGB color space and HSV color space. The SVM (Support Vector Machine) classifiers are trained off-line before labeling procedure. We evaluate our approach on real-world traffic surveillance datasets, demonstrating the effectiveness and the robustness against the outdoor environments. The on-line moving objects labeling is implemented in traffic surveillance videos, including type labeling and color labeling.
Access is limited to the University of Missouri--Columbia.