Tracking objects using image disparities
Image and Vision Computing - Special issue: 5th Alvey vision meeting
Image and Vision Computing - Special issue: 5th Alvey vision meeting
Visual tracking of known three-dimensional objects
International Journal of Computer Vision
Robust model-based motion tracking through the integration of search and estimation
International Journal of Computer Vision
Efficient Region Tracking With Parametric Models of Geometry and Illumination
IEEE Transactions on Pattern Analysis and Machine Intelligence
Robust Tracking of Position and Velocity With Kalman Snakes
IEEE Transactions on Pattern Analysis and Machine Intelligence
| Hi-index | 0.00 |
Visual tracking is essential for many applications such as vision-based control of intelligent robots, surveillance, agriculture automation, medical image processing, and so on. Especially, a fast and reliable visual tracking is important since the performance of visual tracking determines the reliability and real-time characteristics of overall system. It is not easy in visual tracking, however, to estimate the configuration of a target object in real-time when the three-dimensional pose of the target object is changing. On the contrary, a human being is able to track an object without the estimation of three-dimensional pose of the object even though three-dimensional rotations and /or occlusion by other objects change the original image of the object. This paper proposes a fast and reliable SSD-based visual tracker insensitive to three-dimensional rotation of an object as well as translation, two-dimensional rotation, scaling, and shear of an object by proposing a performance measure for distortion of current image with respect to an original reference image. The performance measure is a combination of aspect ratio of a rectangle, variations of four internal angles of the rectangle from ninety degrees, the direction of rotation and the angular velocity of the rectangle. So, the reference image for visual tracking is updated whenever the performance measure is greater than an initialized distortion rate without the estimation of three-dimensional pose of an object. The algorithm is experimented in real-time successfully at a personal computer adopted with a general-purpose frame grabber.