Digital Signal Processing Reference
In-Depth Information
By fusing visible and thermal infrared channel information, the integration of com-
plementary information [4] is available to achieve moving targets detection quickly and
accurately. Here, we propose a multi-phase visible-infrared image fusion strategy for
motion detection. An improved KIRSCH algorithm is used to obtain the complete out-
line of the prospects in experimental materials. Also, the complementary characteristics
of the visible and infrared images are obtained by use of the channel-replacement-
operation. This method can not only effectively remove the shadow of the foreground in
visible image, but also access to the clear and complete moving targets.
2
Method
2.1
Algorithm Framework
The algorithm processes framework as shown in Figure 1. First of all, after data regis-
tration [5, 6], the VI (visible image) and the IF (infrared image) are used to obtain the
foregrounds respectively. Then these foregrounds are fused for new foregrounds, we
call it the FFI (fused-foreground-image). At the same time, the original VI and IF are
also integrated into a new image named FI (fused-image). Then, the FI, the FFI and
the intensity channel of IF are input to an improved KIRSCH algorithm, and with the
help of some image morphological operations, it turns out to be the complete targets.
Fig. 1.
Flowchart of the algorithm
2.2
Foreground Detection
With clear texture and color, frames differential is the best choice to get a more ob-
vious contour for visible frame sequences. Since infrared images have complementary
properties, and with a higher precision and recall rate [8]. GMM (Gaussian mixture
model) [9] can extract the moving targets in a good performance. Each pixel value in
infrared image to obey a certain distribution in the time axis, the distribution can be
representing as a weighted superposition of K Gaussian distribution. For example, a
pixel value at time t is expressed as
X
,
which is a three-dimensional vector, then
:
the distribution of
X
credited as
PX
PX
∑ w
,
ηX
µ
,
Σ
,
(1)
