Development of Visual Tracking Control and Visual Odometry from Dynamic Information Acquired through Sequential Images
Date Issued
2011
Date
2011
Author(s)
Ding, Jun-An
Abstract
Visual tracking is to maintain the target at a specific location in the image by moving the camera. Usually, the specific location is the center of the image frame. At first, the target position is detected by image processing in the image frame. Then, the camera is moved to maintain the target in the image center. This way, visual tracking is achieved. However, if the target is dynamic and the system is not predicting target motions, the target will not be in the image center. The target position will always be off from the image center. This offset is called image error. The image errors caused by target motion is called tracking delay. Therefore, this thesis proposed four motion models to predict the target motion in order to overcome the tracking delay and reduce the image error. The experimental results show that the tracking performance of a predictive system is better than a non-predictive system.
Research on autonomous navigation has been more popular in recent years. One of the autonomous navigation issues is accurately estimating a vehicle’s global location. Therefore, by using visual odometry to estimate a vehicle path, a vehicle’s global location can be found.
The aim of visual odometry is to estimate a vehicle path from captured images. A camera is mounted on a vehicle, and then the camera captures images when the vehicle is moving. Finally, visual odometry estimates the vehicle path from detecting the differences in captured images. However, the standard camera is a projective sensor; it detects only the image change but not the actual camera motion. Thus, this thesis detects the pixel displacement on a special location by using mean absolute error method. Then, the vehicle path is estimated by converting pixel displacement to camera motion. The experimental results show the estimated vehicle path is close to the actual vehicle path by using the proposed method.
Subjects
Visual tracking
image error
tracking delay
pan tilt unit
visual odometry
absolute mean error
post-processing
Type
thesis
File(s)![Thumbnail Image]()
Loading...
Name
ntu-100-R98921009-1.pdf
Size
23.32 KB
Format
Adobe PDF
Checksum
(MD5):2d25815a41930db9fc373b26ddaad012