Advances in Visual Tracking
Instructor: Ming-Hsuan Yang, UC at Merced
Motivation
Recent years have witnessed significant developments in visual tracking with robust and real-time performance. Many novel research algorithms and commercial products have been proposed for numerous applications including video analysis, video conferencing, surveillance, human-computer interaction, action recognition, among others. This course aims to bring the image processing community a review on the state-of-the-art algorithms and systems in visual tracking with emphasis on robust and real-time performance. Applications and demonstrations of these algorithms will be presented. Potential research problems will be identified and discussed at the end of this course.
Course Description
Research on visual tracking covers a wide range of topics, and the focus of this course focus on object-level visual tracking. At the beginning of this course, I will discuss the taxonomy of vi- sual tracking and set up the right stage. The fundamental issues of visual tracking: representation, motion model, and search strategy will be discussed in depth. As visual tracking is related to numer- ous topics in computer vision, the most relevant research areas, e.g., object detection, recognition, background modeling, and segmentation will be discussed in the context of tracking. The recent trend of visual tracking has been on tracking by detection, and several representative works will be presented. Numerous tracking algorithms have been proposed, but considerably less attention has been paid to performance evaluation. To advance the state of the art in this field, important issues in assessing these algorithms will be examined. This short course will be concluded with discussions on applications, demonstrations, and comments on potential research problems.
This course will cover the following topic:
- Taxonomy: feature tracking, object tracking, non-rigid object, articulated object, multiple objects
- Representation scheme: edge, contour, template, blob, color, texture, appearance model, filter response
- Motion model: Euclidean motion, similarity transform, affine transform
- Search strategy: greedy search, random walk, Kalman filter, particle filter
- Related topics: object detection, object recognition, background modeling, segmentation
- Tracking by detection
- Performance evaluation: metric, criteria, ground-truth data
- Applications
- Demonstrations
- Future research directions