Digital Signal Processing Reference
In-Depth Information
Fig. 5.11
Overview of multiview video tracking approaches
Tabl e 5. 1
Feature-based tracking approaches
Ref. Features Tracker Rep.
IT NF [
3
] Color, depth, motion No FM
FCV [
27
] Color, height, width, stereo Kalman filter BB
[
49
] Shape, appearance, depth, motion Kalman filter BB
[
40
] Appearance, depth Bayer tracker BB
FCT [
29
] Luminance, color, human template, 2D position Kalman filter BB
[
43
] Stereo, 2D position, intensity Kalman filter FM
CT [
21
] Color histogram, 4D entities of rectangle Particle filter BB
[
35
] 5D state space of ellipse Particle filter EM
IT
: Independent tracking;
CT
: Cooperative tracking;
NF
: No fusion;
FCV
: Fusion to common
view;
FCT
: Fusion to common tracker;
Ref.
: Reference;
Rep.
: Representation;
FM
: Foreground
mask;
BB
: Bounding box;
EM
: Ellipse model
5.3.1
Feature-Based Tracking
Feature-based tracking methods employs feature match framework with two steps:
feature extraction and feature matching. Feature-based tracking can be performed
either independently in each view or cooperatively across views. A classified sum-
mary of feature-based tracking approaches is presented in Table.
5.1
.
5.3.1.1
Independent Tracking
Independent tracking firstly implements single camera detection, segmentation, and
tracking on its own view. Based on whether the multi-camera fusion module is in-
volved or not, the independent tracking is further divided into
independent tracking
without fusion
and
independent tracking with fusion
.