Smart Video Surveillance System

Theft detection (specifically GUN detection) is one of the most problematic crime that citizens face in their everyday life be it at homes, railways, airports or other public places. Since traditional approaches have long since failed to address this issue there is a dire need of a smart surveillanc

Project Title

Smart Video Surveillance System

Project Area of Specialization

Computer Science

Project Summary

Theft detection (specifically GUN detection) is one of the most problematic crime that citizens face in their everyday life be it at homes, railways, airports or other public places. Since traditional approaches have long since failed to address this issue there is a dire need of a smart surveillance system that is able to distinguish among actions recorded in real time and see if there is any such activity occurring, would generate alarming condition. The proposed system does not waste its memory by recording the activity unnecessarily as it detects on real-time. Hence saves lot of wastage of memory of hard disk.

Project Objectives

The main objective of this project is to make the system active in terms of detecting the malicious activity that includes gun should be detected in an automated manner. It differs from the normal CCTV systems that they caught any activity but can’t be able to detect it on run time, someone is required to watch all the captured frames and then select one frame in which the activity was carried. This feature makes this system special from other that if it will detect the gun in the monitoring area then it will detect it and generate the alarming situation. Generating the alarms after the gun detection also included in the object of this Surveillance system.

Project Implementation Method

During the implementation of the project, we have searched for several different algorithms which provide the best efficiency for our system. The main purpose of this research based project is to come-up with the idea that gives best productivity and less execution time. The project’s accuracy does not depends upon only on the implemented algorithms but the major part is count-on the result which the hardware systems like cameras provides to the system.
The algorithms that we have shortlisted for our project perspective are widely used in Video Surveillance system. These are Convolutional Neural Network (CNN) and You Only Look Once (YOLO) etc.

The Convolutional Neural Network (CNN) is a deep learning algorithm which takes image as input and apply some observations on the particular aspect/part of images. The pre-processing required in this algorithm is much less as compared to many other algorithms. CNN based upon some filters which provide a better Spatial and Temporal Dependencies. Hence, it will trained the image better. Filters are used in ConvNet to detect spatial patterns i.e. edges and lines of particular given image by detecting the fluctuations in intensity value of the images. The ConvNet split the images into much smaller parts which provides a better predictability for the image without losing any distinct feature.
The working of CNN is defined as it takes image as a single vector, then the layers of CNN apply some functions and filters to the image for getting the better trained result, it is called Hidden Layers Functionality. The neurons are arranged in 3 dimensions that are width, depth and height. The output layer provided the result that was required by the user as trained image.
Through CNN we are able to get the accuracy between 52%-60%.

Yolo “You Only Look Once” is an object detection algorithm. Yolo is basically based on regression principle. It process 45 frames per second. It does not only consider the interest part of the image although it predicts the classes and bound box of the whole image.The algorithm see the images in both training and testing time so it will be able to completely encode contextual information about the classes. YOLOv3 is a further enhancement in YOLO with more improved results and much better than other detection algorithms. The accuracy for small objects are improved with the third version of YOLO i.e. YOLOv3 as previous version was not providing satisfying results for small objects.
By implementing YOLO we have achieved the accuracy between 80%-85%.

Benefits of the Project

There are so many benefits of this project that makes it prior to other surveillance systems. Firstly, the system will detect the malicious activity that includes gun in it. No monitor controlling person is required to watch the recordings of the system and detect the gun from it on run time. The system is developed in a manner that it will detect the gun automatically in the premises of the monitoring area. Other CCTV systems are required to have one person on the monitor which operates it manually but this system need no person on the monitor rather than doing the job on its own. The system will generate an alarming situation where the system is located and it will alert all the present staff/people near by the monitoring area. This will help to control the present situation in a better way than calling the officials after the scene. In technical way, it saves lot of memory by just saving the frames in which the activity (includes gun) was detected. The system will be deployed as it is where the requirement is. This system can be deployed in malls, banks, offices, streets etc.

Technical Details of Final Deliverable

One of the main final deliverables is to train the dataset that are not of very fine quality as we did in the FYP 1 deliverables. The images that are not of very good quality like images of CCTV systems, some blurred images in which the gun is not purely visible (concealed weapons). This will be done by using CNN and YOLO. Both models are very efficient in some other ways, YOLO is a predefined model which detects the object with a high accuracy as compare to CNN. CNN is a deep learning algorithm which takes the interest region of the image and apply filters and convolutional layers to it for pre-processing, training and then produced the required result. Through CNN and YOLO we will be able to achieve the better accuracy in terms of detecting the gun.

Final Deliverable of the Project

Software System

Core Industry

Security

Other Industries

Core Technology

Artificial Intelligence(AI)

Other Technologies

Sustainable Development Goals

Industry, Innovation and Infrastructure

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com