2. NoxEye: An AI based threat detector for
Intelligence Surveillance and Alert System
3. Objective
• The main objective of this project is to provide crime detection and
proactive alerts using edge-and fog-integrated approaches.
• To provide intelligent applications with AI.
• Provides an alert by sending the crime data instantly to the police or
protective service, and thus, it ensures a quick response.
4. Abstract
• Nowadays, theft has become a major issue to be dealt with.
• It is not possible for a human to monitor for several hours continuously.
• To overcome the complication in thefts, surveillance cameras are used.
• The main ability of surveillance camera is to just record the act of theft from
which only the knowledge of the intruders will be gained.
• This project presents NoxEye, a lightweight AI-powered threat detector for
intelligent surveillance cameras, which can be deployed on-site at the edge.
5. Existing System
• Framing and Blob recognition (FBD) for Input video preparing and (HON)
Human tracking.
• HOG is in charge of pulling back shape data of question in picture utilizing
force angles and edge headings.
• General purpose computing on graphics processing units (GPGPU) that is
appropriate for use in human stance estimation, and accomplishes constant
execution. A diminishing calculation iteratively expels limit voxels from a
question create a topologically identical skeleton.
6. • K-nearest neighbor is a strategy for grouping objects in view of
nearest preparing cases in the component space.
• Scale Invariant Feature Transform (SIFT) has been turned out to be the
most powerful nearby invariant element descriptor.
• SVM kernel functions based classifiers to distinguish the malicious
nodes from benign ones via evaluating the variance in their Driving
Pattern Matrices (DPMs).
7. Disadvantages
• The project or system was unable to detect motion in case of moving
camera.
• The risk of outliers and ghost points, reduces the performance of the
motion detection system.
• Target harden to find
• Reduce provocation
• IoT-cloud architecture has issues regarding bandwidth, energy, and
latency in real-time video surveillance applications.
• Delayed multimedia IoT tasks.
8. Proposed System
• This paper presented the design and implementation of Nox-Eye, an
AI-enabled threat detector for realtime video surveillance.
• With the help of AI, each fog node can detect and identify a possible
crime event and crime object by processing the motion-captured
images sent by an edge node.
• FRCNN model running at a fog node detects and labels the images
with the name of the crime objects having the highest probability, and
saves those images.
9. Advantages
• The proposed system is far more efficient even without the video
compression algorithm.
• Proved the superiority of our proposed system in terms of agility,
scalability, energy, and CPU and memory usage.
• Efficient crime predictive system
• Real-time crime event detection, ensuring resource efficiency and
good distribution of the processing load in an IoT-based video
surveillance system.
10. System Architecture
Crime Data
Set
Robbery Data Set acquisition
Preprocessing
Action Detection
Feature Extraction
Classified Result
Matching
Police Station
Crime
Database
Server
Classification
Classified Result
Live Crime
11. Modules Split Up
1. NoxEye Control Panel
2. Threat Detector Learning Phase
2.1. Suspicious Video Annotation
2.2. Frame Extraction
2.3. Preprocessing
2.4. RNN Object Detection
2.5. Feature Extraction
2.6. FRCNN Classification
3. Threat Detector unit
3.1. Live Video Annotation
3.2. Threat Detection
4. Custom Notification
5. Performance Analysis
13. Expected Outcome
• High prediction accuracy and response time.
• System would create a better opportunity for security personnel to
detect various types of weapons in real-time.
• Prevent a potential crime.
• A user-friendly interface was developed on top of both models to
allow users to interact with the system conveniently at the camera and
cloud sides.
• Motion detection module that can detect moving objects in
surveillance videos in realtime.
14. References
1. G. F. Shidik, E. Noersasongko, A. Nugraha, P. N. Andono, J. Jumanto, and E. J. and Kusuma,
``A systematic review of intelligence video surveillance: Trends, techniques, frameworks, and
datasets,'' IEEE Access, vol. 7, pp. 457-473, 2019.
2. J. Lim, M. I. Al Jobayer, V. M. Baskaran, J. M. Lim, K. Wong, and J. See, ``Gun detection in
surveillance videos using deep neural networks,‘’ in Proc. Asia Pacific Signal Inf. Process.
Assoc. Annu. Summit Conf. (APSIPAASC), Nov. 2019, pp. 1998-2002.
3. Y.-X. Liu, Y. Yang, A. Shi, P. Jigang, and L. Haowei, ``Intelligent monitoring of indoor
surveillance video based on deep learning,'' in Proc. 21st Int. Conf. Adv. Commun. Technol.
(ICACT), Feb. 2019, pp. 648-653.
4. S. Ren, K. He, R. Girshick, and J. Sun, ``Faster R-CNN: Towards realtime object detection with
region proposal networks,'' IEEE Trans. Pattern Anal. Mach. Intell., vol. 39, no. 6, pp. 1137-
1149, Jun. 2017.