Multiview Image Segmentation and Video Tracking - Video Segmentation and Its Applications - page 136

Fig. 5.13 Detection and tracking using multi-layer homography: projection from multiple views
to several parallel planes

using high elevated cameras. Multi-height homography of three planes parallel to
the floor is involved to detect the head top of the people, which is the feature point
to track the human motion.

5.3.3 Proposed Multiview Video Tracking Algorithm

Unlike the approaches only tracking the interest points (feet, head) of the target,
we propose a feature-based tracking method that define the entire object region as
the tracker and update the deformable objects' regions frame by frame with various
dynamic motions.

5.3.3.1 Tracking of Multiple Separated Objects

Segmenting the consecutive frame is achievable as the motion information is known.
Motion prediction is a form of tracking, which enforces the temporal consistency
between adjacent frames in video. The coarse prediction of the current frame is pro-
jected by pixel-based motion compensation from the mask of its previous frame.
Uncertain band construction and validation for individual object as in Sect. 5.2.3.2
results in the projection refinement. The excellent performance of the tracking strat-
egy has been demonstrated on a couple of real and complex videos containing
spatially separated objects as illustrated in Fig. 5.14 .