|Monitoring Human Activity|
A project of the
Artifical Intelligence, Robotics and Vision Laboratory
University of Minnesota, Department of Computer Science and Engineering
Monitoring Crowded Scenes
Monitoring crowded urban environments is a vital goal for many of today's vision systems. Knowing the size of crowds and tracking their motion has many applications. For example at traffic intersections, intelligent walk-signal systems could be designed based on the number of people waiting to cross. Also, the knowledge of the number of people walking through a crowded area, e.g., outside a school or outside the premises of a public event can be helpful in planning urban environments, general safety, and crowd control. We estimate accurately the counts of people in a scene without constraining ourselves to individuals. This includes dense groups of people moving together. We do this in real-time and place no constraints as far as camera placement or about the size of the groups as far as number of people.
P. Kilambi, O. Masoud, N. Papanikolopoulos, "Crowd Analysis at Mass Transit Sites", IEEE International Conference on Intelligent Transportation
Systems, pp. 753-758, Seattle, WA, Sep. 2006.
View-Dependent Human Motion Recognition
In this project, we attempt to classify human motion into one of several classes. The methods we developed use motion features directly rather than try to reconstruct 2D or 3D models of the human body. We then use Principle Component Analysis for training and classification. In the experiments, we use a data set consisting of 232 video sequences (29 people, each performing 8 different actions).
Online Motion Recognition
In this project, we use a motion recognition strategy that represents a videoclip as a set
of filtered images, each of which encodes a short period of motion history. Given a set of
videoclips whose motion types are known, a filtered image classifier is built using support vector machines.
In offline classification, the label of a test videoclip is obtained by applying majority
voting over its filtered images. In online classification, the most probable type of action at
an instance is determined by applying the majority voting over the most recent filtered
images, which are within a sliding window. The effectiveness of this strategy was demonstrated on real
datasets where the videoclips were recorded using a fixed camera whose optical axis is
perpendicular to the person's trajectory.
View-Independent Human Motion Recognition
Here we study the use of image-based rendering to generate optimal inputs to an entire class of view-dependent human motion recognition systems. Orthogonal input views can be created automatically using image-based rendering to construct the proper view from a combination of non-orthogonal views taken from several cameras. This allows the systems to robustly recognize motion taken from any angle, making their real world application more viable.
R. Bodor, B. Jackson, O. Masoud, N.P. Papanikolopoulos, "Image-Based Reconstruction for View-Independent Human Motion Recognition", IEEE International Conference on Intelligent Robots and Systems IROS2003, pp 1548-1553, Las Vegas, Oct. 2003.
R. Bodor, B. Jackson, , O. Masoud, N.P. Papanikolopoulos, "Image-Based Reconstruction for View-Independent Human Motion Recognition," Technical Report, Artificial Intelligence, Robotics and Vision Laboratory, Dept. of Computer Science and Engineering, University of Minnesota, Mar. 2003.
Monitoring Bus Stops
G. Gasser, N. Bird, O. Masoud, N.P. Papanikolopoulos, "Human Activities Monitoring at Bus Stops", IEEE International Conference on Robotics and Automation ICRA2004, Apr. 2004.
General Activity Recognition
The protection of critical transportation assets and infrastructure is an important topic these days. Transportation assets such as bridges, overpasses, dams and tunnels are vulnerable to attacks. In addition, facilities such as chemical storage, office complexes and laboratories can become targets. Many of these facilities exist in areas of high pedestrian traffic, making them accessible to attack, while making the monitoring of the facilities difficult. In this research, we developed components of an automated, "smart video" system to track pedestrians and detect situations where people may be in peril, as well as suspicious motion or activities at or near critical transportation assets. The software tracks individual pedestrians as they pass through the field of vision of the camera, and uses vision algorithms to classify the motion and activities of each pedestrian. The tracking is accomplished through the development of a position and velocity path characteristic for each pedestrian using a Kalman filter. With this information, the system can bring the incident to the attention of human security personnel. In future applications, this system could alert authorities if a pedestrian displays suspicious behavior such as: entering a "secure area," running or moving erratically, loitering or moving against traffic, or dropping a bag or other item.