Real Time Object Tracking On Active Pan Tilt Zoom Camera Using Cmt

Open Access
Davis, Kameran J
Graduate Program:
Computer Science and Engineering
Master of Science
Document Type:
Master Thesis
Date of Defense:
April 01, 2015
Committee Members:
  • Vijaykrishnan Narayanan, Thesis Advisor/Co-Advisor
  • John Phillip Sustersic Jr., Thesis Advisor/Co-Advisor
  • Jack Sampson, Thesis Advisor/Co-Advisor
  • CMT
  • Keypoints
  • Tracking
  • Real-time
  • Pan-Tilt-Zoom
  • Camera
  • Consensus-Based
  • Matching
  • Software
  • Canon
  • VB-C300
Object tracking is a fast growing topic in computer vision. Object tracking is especially useful in terms of unmanned and manned aerial, ground, and underwater vehicles, making inroads to replace manned vehicles for purposes such as videography, structural monitoring, surveillance, and aerial mapping. Other applications include motion-based detectors as well as human-computer interaction. In this thesis, a feedback-based solution to automatically track a selected target is proposed and implemented, based on object motion in different parts of the scene and experimentation is performed to illustrate fulfillment of the Pan Tilt Zoom operations of the camera while tracking test subjects in real-time. The approach uses Consensus Based Matching and Tracking of Keypoints to track the centroids of the objects. After each frame, the algorithm updates the location of the object centroid relative to the center of the camera’s point of view. With the proposed real-time tracking system, this location is then used to calculate what direction and by how many degrees does the camera need to move so that the object of interest is contained with the camera’s center of view. This solution can be even be extended to function on multiple cameras tracking a single object, allowing to factor in other elements such as depth perception, or even the expanded to function as part of a fully autonomous localization and navigation systems for unmanned vehicles and robots.