Video Based Human Activity Detection
Departments: Computer Science and Engineering Advisor: Dan Lopresti
Smart spaces - buildings or areas with ambient intelligence - should be able to detect the actions and behaviors of their inhabitants unobtrusively, with as little explicit querying as possible. Typical human actions, however, are complex, and difficult to capture accurately. Using real-time video capture and analysis techniques leads to better, more accurate, more robust computational techniques for the detection of the presence and actions of humans, resulting in more effective smart space implementations. This research focused on using data from video streams to detect and classify the presence, location, motions, and actions of people in real time within the Smart Spaces section of Mountaintop Bay C. An AXIS M1004-W wireless network camera was used to capture and send video to a C++/OpenCV program for analysis. Using background subtraction, blob-extraction and filtering, optical flow calculation, and blob velocity calculation, this program was able to detect the motion and estimate the relative velocities of people within the space in real-time, with relatively low latency. Action detection, while not implemented during the research period, could be achieved by training classifiers to recognize actions based on the relative velocities of different body parts.
About Connor Tench:
Connor Tench is a senior majoring in Computer Science (B.S.) from Allentown, PA. His research interests include computer vision, pattern recognition/machine learning, bioinformatics, and artificial intelligence. Outside of academic work, Connor is the manager of the Lehigh University Choir, a group which performed at Carnegie Hall in New York during November 2014, and which will perform a weeklong tour in Italy this May. He oversees the day-to-day running of the group, and is working to streamline its administrative policies to allow for continued sustainable future leadership. Additionally, Connor is a Technology, Resource, and Communications (TRAC) fellow. Each semester he works with a group of students in a single class by reading first drafts of their writing assignments, offering commentary on their writing, and conferencing with each student individually to discuss future steps in their writing.