150 likes | 172 Views
A presentation discussing an adaptive model for tracking objects by color, including background, results, procedure, and analysis of the work. The relevance of the work to different methods of color tracking algorithms and blob tracking methods is also discussed.
E N D
A presentation of “An Adaptive Model for Tracking Objects By Color” By Chris Rasmussen, Gregory Haeger Dept. of Computer Science Yale University Presented by Paul Alan Roberts
Outline • Purpose of Paper • Background • Results of their work • Procedure and concepts • Their analysis of their work • My analysis of their work • Summary of relevance of their work
Purpose • To compare and contrast various methods of color tracking algorithms • Give detailed discussion of blob tracking • Comparison of different blob tracking methods
What is color tracking? CT is the science of teaching a comp-uter what an object is, and how to main-tain its position as it moves within the field of view of a camera. Photo courtesy of Yale University, Dept of CS
Background • 3 basic image processing techniques used for color tracking. • Edge detection • Variations on region-based correlation • Image segmentation
What is blob tracking? • Blob-tracking is concerned with finding the center of mass of a image, the blob. • The blob is defined as the region of interest in an image, and is usually extracted using a fast color segmentation technique to allow for real time processing. • Generally blobs are tracked in small windows to reduce the amount of computations necessary.
Blob Tracking • Difficulties Associated with Blob Tracking • Unknown backgrounds and color distribution • Real time tracking limits computations complex • Nearly complete elimination of false positives • Backdrops are not always suitable or practical. • The more accurate we make our algorithm, the longer it takes to process. • Shifts in lighting combined with noise can cause the tracking program to lose track of the object.
Results Video courtesy of CIRL at Johns Hopkins University
Initialization • Program needs to know the amount of color variance. • All possible lighting situations should be sampled. • Avoid over saturated pixels.
Methods • SC :Finds the center of the image and creates a small window around it. • AC: Tracks color variation. Tracks the shift in mean color distribution. • ADC: Removes noise by requiring large amounts of pixels to be present to constitute the image. Requires that the object have a large amount of pixels of the same color in a given region.
Author’s analysis • Static Color: model is best suited for tracking objects with static predicable backgrounds • Adaptive Color: suitable for tracking objects with varying light intensities and shadowing. • Adaptive Distraction Resistant: most useful algorithm especially when there are periphery pixels similar in color to the target.
Presenter’s Analysis • Thorough discussion of how to recapture an image once it has been lost is lacking. • SC analysis: very useful for limited field studies such as a paddle ball playing robot. Low chance of losing the object and high frame rate for tracking achievable.
Presenter’s Analysis (cont) • AC has difficulty when tracking object that experience a large change in color space. • ADC has trouble tracking an object when it’s size changes. • All three methods have their distinct advantages and disadvantages. ADC has the most uses but is the least robust while SC is highly restricted but very robust.
Relevance to our work • Blob tracking will allow the user to move the board within the field of view of the camera, while maintaining a lock on the image’s coordinates. • Opens the door to adding features such a touch buttons, hand panning, and fist zooming.
Conclusion • Blob tracking is an efficient, straightforward way to track an object in the field of view of a camera. • It has a myriad of practical uses. • It fulfills the requirements for our project