Abstract:
The increase in terror attacks globally is giving rise to the requirement of early
detection and positioning of weapons. The system has proposed approaches that
able to classify armed weapons that are visible in an image and are in attacking position.
This work presents an armed weapon classification system based on their directions
with in images which is appropriate for both, surveillance and control purposes.
This research focused on limiting the number of incidents that take place by carrying
armed weapon like pistols and rifles. It also managed to specify the positions of
pistols and rifles whether it is attacking position or not, by evaluating the linear and
perpendicular positioning of arm.
are
an
In this report, an automated surveillance system for classifying armed weapons and
their attacking and non-attacking situations in cluttered scenes is proposed. To obtain
this purpose, we have divided it into two approaches; one approach is for weapon
classification and second is for positioning of weapon. The first approach is SIFT
combined with SVM and the second is CNN based approach. For the first approach
SIFT key-points are extracted from an image then the images are divided into clusters
by applying k-means clustering then, histogram is implemented, then the images
classified by using SVM for classification based on the histogram. For our second
approach, CNN architecture Inception-v3 is used to get better performance, which is
done using Tensorflow
