In Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy
will defend his dissertation proposal
Human Groups as Contextual Feature in Video Analysis
With a growing network of cameras have being used for security applications, video-based monitoring relied on human operators is ineffective and lacking in reliability and scalability. In this proposal, we develop automatic solutions that able to analysis the human-related information in the video, such as identifying same persons across different cameras (human re-identification) and recognizing the human activities.
Analyzing video only using individual-based features can be very challenge be cause the significant appearance and motion variance due to the changing of viewpoints, different lighting conditions, and occlusions.Motivated by the facts that people often form groups in the scene, it shows that the interaction among group members can dis-ambiguous the individual features in video analysis tasks. This proposal introduces two descriptors to leverage the human group as contextual information. Subject Centric Group (SCG) feature captures person group's appearance and shape information using the estimation of persons' positions in 3D space. The metric is designed to consider both human appearance and group similarity. Spatial Appearance Group (SAG) feature extracts group appearance and shape information directly from the video frames. A random forest model is trained to predict the groups similarity score. This proposal demonstrates the application of proposed features in person re-identification problem. The experiments show that proposed features can reach state-of-the-art accuracy on challenge re-identification dataset that represents the real-world scenario.
Date: Wednesday, May 4, 2016
Time: 10:00 AM
Place: PGH 501
Advisor: Prof. Shishir Shah
Faculty, students, and the general public are invited.