Smart Video Surveillance System

Project Title

Smart Video Surveillance System

Project Area of Specialization Artificial IntelligenceProject Summary

For security reasons, video surveillance systems are being used everywhere, however these systems are neither intelligent nor automated to address currently increasing security issues.

Few of these issues identified in this project are Browsing Time Problem as you have to browse complete video to find any specific moment in the video which requires huge time and efforts. Monitoring Problem as you need to hire a person to keep a watch on the streaming videos every time. No on-time notification of object detection. Surveillance systems are passive as they do nothing but recording twenty-four by seven (24/7). Cost of Storage as a video recording at rate of 30fps normally takes up to 72.21 GB per day.

To address above issues, we propose a system “The Smart Video Surveillance System (SVSS)” which is a desktop-based application that is smart and autonomous while having a clean and user-friendly interface for ease of use by the users. It introduces the improvements and optimization in existing systems while solving the major issues already mentioned.

This project optimizes the recording time and storage by recording only when a certain specified object is detected inside video frame such as a car moving around in a restricted area and the system is in surveillance for (24/7) Twenty-Four by Seven while the system is working on object detection and classification mechanism, thus there is no need for a person to sit and monitor the video streams, the system will automatically send the notification on the users’ cell phone when a certain specified object is detected.

This system works by processing video streams or frames at real-time and for that, the system uses a pre-trained model of classification that is trained using deep learning techniques. The system makes use of Multi-Class Classifier, Deep Learning and Computer Vision techniques along with different libraries, and algorithms to detect, classify the objects and respond accordingly. 

The system provides its users the capability of selecting the type of objects to detect, classify and then either send a notification to the user or/and record video if and only if that specified type of object is detected and then the user can take measures accordingly.

The final product will provide the benefits to the user such as; browsing time of the video will be optimized, the system will be autonomous, it will automatically send the notification to the user when a specified object is detected, storage of the recorded videos will be optimized and cost of buying the storage devices will be saved.

Project Objectives

The researchers have proposed smart surveillance systems with additional features for more accurate monitoring of events, but much attention is paid to make an efficient system that utilizes minimum resources. This system is going to make the changes in conventional surveillance systems and add the multiple different capabilities like; Object detection, On-time Notification and Video Recording (Up-on the specified object detection)

The system will address these problems by providing the system with the capability to intelligently monitor the objects appearing in front of the camera and make the decision of recording based on the judgment criteria specified by the user. In this way, only a specific type of object will be recorded and rest frames will be ignored by the system. Discarding the unconcerned frames will cause to reduce video size which will impact both storage as well as cost reduction.

The system smartly monitors the video frames and identifies objects and notify the user when any user-specified object is detected which will provide some relaxation to the person monitors streaming. Last but not the least problem of scanning time. In the existing passive system, the entire recording is scanned/browsed to find any specific event/object in the saved video. The system provides solution to this problem by recording only concern frames which will save scanning/browsing time and the event in which the user is interested can be found with minimal efforts.

The end product would be a system that is Smart and autonomous. It won’t need a person for most of the operations.

Project Implementation Method

To implement this project there are many methods that will be involved in the process:

1. For Training the model for the classification system, the existing datasets will be explored, then the system will be trained on the image dataset such as COCO dataset. The Multi-Class classifier (YOLO V3) will be used to classify Video frames at runtime.
2. For the Realtime Object Classification, the system will capture the video frames at runtime and classify them based on a trained classification model by using the Deep Learning techniques or algorithms.
3. For Recording the video, the frame differencing will be used for detecting the motion and then the classifications algorithms will be used to classify the objects in those frames according to the user given criteria.
4. For the Notification service, the ‘Twilio’ python library will be used that will send the notification to the users’ cell phone upon the detection of the specified object in frames

Benefits of the Project

There are many benefits that this project provides, such as;

1. The Storage of the recorded videos will be optimized as we will only record the video frames when a user-specified object is detected.
2. The Cost of the storage will be saved, as there will not be a lot of video frames stored and the need for buying the storage devices will be reduced.
3. The browsing time of the video will be saved, as there will only be the recording of those selected frames, which contain the user-specified object.
4. The system will be autonomous as there will be no need for a person to sit and monitor the video streams. The system will automatically send the notification to the user when an object is detected.

Technical Details of Final Deliverable

 Operating Environment
This system will work on common machines people use in their routine. The minimum supported hardware recommended is core i3 with 8GB RAM. Other things included in the operating environment are the following:
•    Operating System: Windows and Linux
•    Runtime monitoring: Camera (Medium level pixels camera is recommended)
•    Program execution: Python (3.7 is recommended)
•    Image pre-processing: Open CV library
•    GUI: Kivy or Python Tkinter Python Libraries
•    Camera connection: Drivers compatible with OS 
System Constraints
1.    Software Constraints
    Our system’s performance is dependent on the quality of the image captured, camera quality, indoor or outdoor conditions and the crowd present in an image. If the required conditions failed to meet for instance illumination present in the image may cause the system’s performance to degrade. Similarly, System may misclassify the image if it does not find image clear or a lot of noise in images such as raindrops.
2.    Hardware Constraints 
    Hardware constraints of this project include not having enough RAM or CPU capability to execute the program and quality of the camera
3.    Environmental Constraints
    Too hot temperature can cause a problem to this system only when running many cameras and storing videos on the server. All required software and libraries should be installed in the system.
4.    User Constraints
    Users should have a good understanding of video surveillance and should be aware of using desktop applications. Users should be able to connect cameras with this system.

External Interface Requirements
    Our system contains the following requirements for its various interfaces.
1.    Hardware Interfaces
    This project intended to use common computer hardware that people use in their daily routine work. Cameras should be compatible with the operating system installed on the computer
2.    Software Interfaces
             Following are the software to be used for Smart Video Surveillance System (SVSS)

Software Used	Description
Software Platform	Desktop based application is chosen because of the high processing task. Users also prefer to browse videos in large screens to find specific information from the video that also influences us to choose this environment.
Keras	Keras is a widely used library for neural networks. This system will use Keras for model training.
Anaconda	Anaconda IDE is chosen for developing the application’s demo. Jupyter notebook will be used for writing code

Software Used

Software Platform

Keras

Anaconda

Final Deliverable of the Project Software SystemCore Industry SecurityOther IndustriesCore Technology Artificial Intelligence(AI)Other TechnologiesSustainable Development Goals Industry, Innovation and InfrastructureRequired Resources

Elapsed time in (days or weeks or month or quarter) since start of the project	Milestone	Deliverable
Month 1	Model training for Object detection	Trained Model
Month 2	System and Interface development	System modules and Interface
Month 3	Integration of all modules	Complete Product (Working Prototype)