Professional Documents
Culture Documents
A SEMINAR REPORT
Submitted by
to
the APJ Abdul Kalam Technological University
in partial fulfillment of the requirements for the award of the
Degree of
Bachelor of Technology
in
Computer Science and Engineering
679582 2022-2023
DECLARATION
Place: Kuttippuram
Date: 09-12-2022 Muhammed Khaleefa Zayed
i
ABSTRACT
We propose a novel life-detection radar using a wideband Boolean-
chaos signal as a probe signal. The range between the radar and the
human target can be obtained by correlating the echo signal from the target
with its delayed duplicate. This range is modulated periodically by human
chest's displacements along the recording time axis and the modulation
frequency is the same as the respiratory frequency. Therefore, the authors
propose a life-detection algorithm based on correlation method to detect the
human target's respiratory frequency and range. Experimental results
demonstrate that the authors’ radar can simultaneously detect the
respiratory frequency and range of the human target behind a 20-cm-thick
wall. In addition, the high range resolution and excellent anti-jamming
property of the chaotic radar have been demonstrated by their previous
studies, which make it perform superbly in complex electromagnetic
environments.
ii
CONTENTS
LIST OF FIGURES
psy chological, and economic cost. Many people die each year from gun-
the short and long terms. Number of studies show that handheld gun is the
primary weapon used for various crimes like break-in, robbery, shoplifting, and
rape.
dataset and thus different from existing patterns. An anomaly is a pattern that
on the phenomenon of interest [3] [4]. Object detection uses feature extraction
classification. Also concerned with accuracy, since a false alarm could result in
adverse responses [11] [12]. Choosing the right approach required to make a
This paper reviewed and categorized various algorithms that have been
used in the detection of handgun and knives with their strengths and
This paper presents a novel approach for weapon classification using Deep
Con volutional Neural Networks (DCNN) that is based on the VGG Net
architecture. VGG Net is the most recognized CNN architecture which got its
problems. Thus, weights of pre-trained VGG16 model are taken as the initial
classes: knife, gun and no-weapon are used to train the classifier. To fine tune
the weights of the proposed DCNN, it is trained on the images of these classes
downloaded from internet and other captured in the lab achieved for weapon
classification.
2
Verma, Gyanendra K. and Anamika Dhillon. “A Handheld Gun Detection using
problem be cause of the various subtleties associated with it. One of the most
frequently. There are two types of occlusions of gun, namely gun to object and
background site/scene structure. Self- occlusion arises when one portion of the
Figure 3.1 shows the methodology of weapons detection using deep learning.
Frames are extracted from the input video. Frame differencing algorithm is
applied and bounding box created before the detection of object [7] [8] [14].
(SSD or fast RCNN) is chosen for gun detection. The approach addresses a
4
CHAPTER 4
ARCHITECTURE
4.1 Faster R-CNN
5
4.2 SSD (Single Shot Detector)
Figure 4.2: Detection and Tracking
region proposal network. To overcome the drop in accuracy SSD brings a few
improvements allow SSD to match the Faster R-CNN’s accuracy using lower
around 74 percent mAP and 59 fps on COCO dataset. Figure 4.2 shows SSD
6
CHAPTER 5
IMPLEMENTATION
5.1 Resources or components used for implementation
3.4.
• Python 3.5- High level programming language used for various image-
processing applications.
7
• Video Size- 4.14 MB
• GPU with high-end computation power were used to remove lag in the
8
5.4 Dataset Creation
(chrome extension) which can download multiple Google Images at once. Then
the down loaded images are labelled. 80% of total images used for training and
20% images for testing. XML data is converted into CSV file by executing the
9
5.5 Training
tf_record file.
• Label map is the file which stores the total number of types of objects that
• The images are converted into tf_record format in Tensorflow so that they
10
5.5.1 Training of Faster R-CNN
11
CHAPTER 6
RESULTS AND ANALYSIS
Figure 6.1 shows detection of a gun AK47 using SSD algorithm. Accuracy of
Figure 6.2 shows detection of COLT M1911 and Smith Wesson Model gun with
12
6.2 Detection of weapons using Faster R-CNN algorithm
Figure 6.3 shows detection of AK47 gun using faster RCNN with an accuracy
Figure 6.4 shows detection of Colt M1911 gun Smith & Wesson Model 10 gun
using Faster R-CNN algorithm with the accuracy of 74% and 91%.
13
6.3 Performance Analysis
The Figure 6.5 shows that highest average accuracy is obtained for pre-
labelled dataset (i.e. AK47) and the Colt M1911, Smith Wesson Model 10, UZI
Model, Rem ington Model obtained accuracy in the range of 76% to 91%.
1.606s/frame. This concludes that the pre labelled dataset provided better
created dataset.
6.3.2 SSD
14
Figure 6.6 shows performance Analysis for the SSD algorithm. Obtained
guns such as AK47, Smith and Wesson Model 10, Colt M1911, UZI Model and
Remington model and obtained maximum confidence level for AK47 gun. Used
SSD and RCNN In ception V2 models to train the guns. SSD took 12 more
hours to train model in com parison with RCNN model but provided lower
provides faster speed than Faster R-CNN by 0.7 seconds. Pre-labeled dataset
like AK47 gun provides higher accuracy in SSD and Faster R-CNN models,
15
CHAPTER 7
CONCLUSION
SSD and Faster RCNN algorithms are simulated for pre labeled and self
created image dataset for weapon (gun) detection. Both the algorithms are
efficient and give good results but their application in real time is based on a
tradeoff be tween speed and accuracy. In terms of speed, SSD algorithm gives
better speed with 0.736 s/frame. Whereas Faster RCNN gives speed
Faster RCNN gives better accuracy of 84.6%. Whereas SSD gives an accuracy
detection due to faster speed but Faster RCNN pro vided superior accuracy.
Further, it can be implemented for larger datasets by training us ing GPUs and
16
REFERENCES
[1] Wei Liu et al., “SSD: Single Shot MultiBox Detector”, European
[2] D. Erhan et al., “Scalable Object Detection Using Deep Neural Networks,”
IEEE Conference on Computer Vision and Pattern Recognition(CVPR),2014.
[5] Abhiraj Biswas et. al., “Classification of Objects in Video Records using
[7] Mohana et.al., “Simulation of Object Detection Algorithms for Video Survil
17
[8] Yojan Chitkara et. al.,“Background Modelling techniques for foreground
Sys tems,2019.
[15] J. Wang et.al., “Detecting static objects in busy scenes”, Technical Report
18
[16] V. P. Korakoppa et.al., “An area efficient FPGA implementation of moving
19