Loading...
| Friend's email | |
| Your name | |
| Your email | |
| enter code | |
This page was sent successfuly
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 40430 (19)
- University: Sharif University of Technology
- Department: Computer Engineering
- Advisor(s): Kasaei, Shohreh
- Abstract:
- Human tracking is one of the most important research topics in computer vision with application in surveillance, crowd analysis, human motion and behavior analysis, and human-computer interaction. Availability of a large number of cameras has caused the need of designing automatic tracking methods. One of the most important challenges in automatic tracking is the occlusion of objects. Several methods have been proposed in the literature to solve this problem. One of the most popular and powerful methods is based on the usage of multiple views of the same scene in a cooperative manner between cameras. The most important challenge in using multiple cameras is the complexity of the tracker in combining camera views and object modeling to reach a unique and consistent labeling for objects seen in all cameras. In this research, a new and efficient method for tracking humen in multiple camera views is proposed. At first, for background subtraction, an adaptive version of Gaussian mixture model is introduced where shadows are also removed by imposing a constraint on the color of background model and observed pixel. Each camera uses a single view tracker by a state machine approach to track the states. It uses the homography relation of the ground plane among different cameras to correspond different tracks obtained by these cameras. Information of tracks is then combined with the ground plane homography relation and color histogram relation of views to obtain consistent tracks. Tracks of each view are then corrected according to the correspondence of total tracks. As there is no need for precalibration, the proposed method can be used in different situations with a least interaction with environment and setting requirements. Experimental results on PETS database shows the superiority of the proposed method over other available tracking methods (such as epipolar and mean-shift) in motion segmentation and track correspondence. Proposed method decreased execution time of motion segmentation algorithm more than 50% in comparison to popular methods such as mixture of Gaussian.
- Keywords:
- Multicamera Tracking ; Human Tracking ; Multiview Tracking ; Occlusion ; Homography ; Uncalibrated Cameras
Digital Object List
-
محتواي پايان نامه
- view
Bookmark