Superpixels shape analysis for carried object detection

Video surveillance systems generate enormous amounts of data which makes the continuous monitoring of video a very difficult task. Re-identification of subjects in video surveillance systems plays a significant role in public safety. Recent work has focused on appearance modeling and distance learning to establish correspondence between images. However, real-life scenarios suggest that the majority of clothing worn tends to be non-discriminative. Attributes- based re-identification methods try to solve this problem by incorporating semantic attributes which are mid-level features learned a prior. In this paper we present a framework to recognize attributes with applications to carried objects detection. We present a supervised approach based on the contours and shapes of superpixels and histogram of oriented gradients. An experimental evaluation is described using a dataset that was recorded at the Greater Cleveland Regional Transit Authority.