3.2. Feature window selection
Before addressing the problem of automatically selecting feature windows, it is necessary to under- stand what qualities we desire the feature window to possess. For our serveing task, we would like feature windows to have the follwing two important qualities:
• The feature window should contain enough pixels that correspond to the object of interest so that the window may be effectively used for tracking. Ideally, each of the feature window’s pixels correspond to the object, causing the window to be invariant with respect to the object’s surround- ings.
• The pixels within the feature window must possess enough contrast for the window to be tracked.
(a-b) The user initially selects a sponta- neous domain representing the beginning of the conveyor belt.
(c-e) As the target appears, its continuous domain initially intersects the spontaneous domain. The two domains are combined into one.
(f-g) As the object moves away, the two domains eventually separate.
Fig. 4. Segmentation domains.
By computing a bounding box around an object of critical because visual tracking algorithms may fail interest, we have provided a solution for the first of due to repeated patterns or areas of uniform intensity these two concerns. The problem of locating in the intensity arrangement. Both situations cause a trackable features for an object has been reduced tracking algorithm to reach ambiguous matches from from a search throughout the entire image to a within its search neighborhood, resulting in an search within the object’s bounding box. output of spurious displacement measures. In order The qualification of feature window intensity to avoid this problem, our system considers and
arrangement with a good confidence measure is evaluates many candidate placements of a feature
window. Our system then selects feature window 3.2.2. Confidence measures. Feature selections made placements that optimize the performance of a from within a bounding box typically lie entirely tracking algorithm. within the object or contain at least a portion of the object’s pixels. Confidence measures are used to
determine which of these are most suited for
3.2. 1. Use of optical flow