Using Bimodal Gaussian Mixture Model-Based Algorithm for Background Segmentation in Thermal Fever Mass Screening
Since the SARS outbreak in 2003, the use of thermal imaging technology for fever mass screening in public areas becomes important. Mass screening is the first stage screening before individual screening for further verification and it is very effective to avoid congestion. In no...
| Main Authors: | , , , |
|---|---|
| Format: | Conference or Workshop Item |
| Language: | English |
| Published: |
2011
|
| Subjects: | |
| Online Access: | http://umpir.ump.edu.my/id/eprint/2112/ http://umpir.ump.edu.my/id/eprint/2112/ http://umpir.ump.edu.my/id/eprint/2112/1/ProcediaS%2BBS_Template_WCIT-2011_belum_upload.pdf |
| Summary: | Since the SARS outbreak in 2003, the use of thermal imaging technology for fever mass screening in public areas becomes
important. Mass screening is the first stage screening before individual screening for further verification and it is very effective to
avoid congestion. In non-contact temperature measurement, the medial canthal - between the corner of the eyes and lachrymal
(tear) duct - is the best area to represent the elevated body temperature for fever detection. The available thermal technology only
provides the ability to detect temperature dimension of objects instead of the regions of interest. For instance, if there is similarity
in temperature between regions of background and medial canthal area, thermal camera alone is unable to identify the correct
region. Hitherto, in most of installed thermal imaging in airports, this problem is only solved by human operators, thus its
effectiveness is influenced by human factors. In this paper, an algorithm based on Gaussian Bi -modal Mixture Models (GBMM)
is proposed for background-foreground segmentation as an important feature to identify medial canthal area. To estimate the bi -
modal background-foreground distribution mixture parameters, Expectation-Maximization (EM) algorithm is applied and the
images are clustered statistically and linearly. The results are later multiplied with background subtraction results to get a better
quality of segmentation. The 640x480 thermal resolution imagery sequences are used as input and the intensity of the pixels are
collected from arriving passengers in Kuala Lumpur International Airport (KLIA) under a controlled ambient temperature.
Twenty image sequences were used in the experiments and the result shown the feasibility of the proposed algorithm.
|
|---|