Suspicious Activity Detection from Video Surveillance Using CNN Algorithm

Abstract

Suspicious Activity is predicting the part or joint locations of someone from a picture or a video. Human suspicious Activity is one amongst the key issues in laptop vision that has been studied for over fifteen years. it\'s necessary due to the sheer variety of applications which might like Activity detection. for instance, human cause estimation is employed in applications as well as video police investigation, animal following and behavior understanding, language detection, advanced human-computer interaction, and marker less motion capturing. Low price depth sensors have limitations like restricted to indoor use, and their low resolution and yelling depth info build it troublesome to estimate human poses from depth pictures. Hence, we have a tendency to attempt to use neural networks to over- return these issues. Suspicious act recognition from police investigation video is an energetic analysis space of image process and laptop vision. Through the visual police investigation, human activities may be monitored in sensitive and public areas like bus stations, railway stations, airports, banks, searching malls, faculty and faculties, parking tons, roads, etc. to stop act of terrorism, theft, accidents and ill-gotten parking, vandalism, fighting, chain snatching, crime and different suspicious activities.

Introduction

I. INTRODUCTION

Human behavior recognition within the globe setting finds lots of applications together with intelligent video police work, looking behavior analysis. Video police work has huge application areas particularly for indoor outside and places. police work is associate integral a part of security. these days security camera becomes a part of life for the protection and security functions.

E-police work is one in every of the most agendas in Digital Bharat, development program of Indian government. Video police work remains as a section of it. Benefits of video police work are effective observation, price effective auditing capability, adopting new security trends etc. With increasing crime rates it becomes a tangle if they're not known in time and necessary perceptional actions taken. Most urban and metropolitan areas have police work systems put in that perpetually accumulates knowledge.

With the huge accumulation of police work knowledge there are higher probabilities of suspicious activities to occur. however these tasks need human oversight to sight such activities as they're too difficult for computing to handle and need high resources. Breaking down difficult tasks associated police work sub tasks that result in potential crimes are a way to change an activity to be machine-controlled. we have a tendency to target 2 main potential ends up in crimes that we have a tendency to arrange to sight through our models.

II. LITERATURE SURVEY

According to [1], The system described then grounded on the principle that when an abnormal event occurs, the most recent frames of videotape will be significantly different than the aged frames. we train an end- to- end model that consists of a spatial point extractor and a temporal encoder- decoder which together learns the temporal patterns of the input volume of frames. The model is trained with videotape volumes consists of only normal scenes, with the ideal to minimize the reconstruction error between the input videotape volume and the affair videotape volume reconstructed by the learned model. After the model is duly trained, normal videotape volume is anticipated to have low reconstruction error, whereas videotape volume confirming of abnormal scenes is anticipated to have high reconstruction error. By thresholding on the error produced by each testing input volumes, our system will be suitable to descry when an abnormal event occurs

According to [2], Several recent approaches propose CNNs that directly break specific imaging problems. These architectures act cases of the ODP frame, still with far different provocations for the design.

According to [3], Sultani et al. used a general model of anomaly discovery using deep shop( Multiple Instance Learning) frame. Frame predicated receiver operating characteristics( ROC) wind and corresponding area under the wind( AUC) are used to estimate the performance of the system. Their experimental results showed that the shop system for anomaly discovery achieves significant improvement on discovery performance as compared to the state- of- the- art approaches.

According to [4], This paper focuses on the problems encountered with the real- world images and how they impact object discovery. The YOLO model is trained using demoralized images, the results reveal that the trained model is susceptible to learning farther features and that it can manage with complex surroundings. Training with demoralized images did much to ameliorate the average perfection of the object discovery, better generality capability and advanced robustness

According to [5], This paper proposes to initiate a mobile application for the detection of abnormal crowd behavior and its management. Server-side application is linked to an IP camera which detects suspicious activity, determines crowd level at the entrance of any public/private place by making use of Social Force Model (SFM) algorithm and sends this information to the user.

According to [6], Some of the early literature involving detection of weapons focused on analyzing x-ray images and infrared images to detect concealed weapons. These systems had One research utilized the color-based segmentation to distinguish objects and used Harris interest point detector and FREAK to detect guns in the segmented images. Even though many methodologies to detect cyberbullying among which deceptive Phishing in IMs for text-related messages being developed

According to [7], There remains some inefficiencies in the detection process of suspicious words in the IM & SNS. As it may also contain shorthand or short-form words in those micro blogs which are being communicated. These words cannot be detected by the help of regular surveillance tools leading to concealment of suspicious messages. Advance Motion Detection (AMD) algorithm was used to detect an unauthorized entry in a restricted area . In the first phase, the object was detected using background subtraction and from frame sequences the object is extracted. The second phase was detection of suspicious activity. Advantage of the system was the algorithm works on real time video processing and its computational complexity was low. But the system was limited in terms of storage service and it can also be implemented with high-tech mode of capturing of videos in the surveillance areas.

III. SOFTWARE INFORMATION

A. Anaconda

What is Anaconda Navigator? Anaconda Navigator may be a desktop graphical interface (GUI) enclosed in Anaconda® distribution that permits you to launch applications and simply manage conda packages, environments, and channels while not victimization command-line commands. Navigator will hunt for packages on Anaconda.org or in an exceedingly native boa Repository. it's accessible for Windows, macOS, and Linux. to induce Navigator, get the Navigator Cheat Sheet and install Anaconda.

B. Spyder

Spyder may be a powerful scientific setting written in Python, for Python, and designed by and for scientists, engineers and information analysts.

It offers a singular combination of the advanced written material, analysis, debugging, and identification practicality of a comprehensive development tool with the information exploration, interactive execution, deep examination, and exquisite image capabilities of a scientific package.

Beyond its several inbuilt options, its talents are often extended even additional via its plugin system and API. moreover, Spyder also can be used as a PyQt5 extension library, permitting you to create upon its practicality and introduce its parts, like the interactive console, in your own software system.

IV. SYSTEM ARCHITECTURE

A. Modules

• Pre-processing
• Feature Extraction
• Classification

In knowledge flow sheet, we have a tendency to Show that flow of knowledge in our system in DFD0 we have a tendency to show that base DFD within which parallelogram gift input likewise as output and circle show our system, In DFD1 we have a tendency to show actual input and actual output of system input of our system is text or image and output is rumor detected likewise in DFD two we have a tendency to gift operation of user likewise as admin.

B. UML Diagram

Unified Modeling Language may be a commonplace language for writing computer code blue prints. UML is method freelance, though optimally it ought to be employed in method that's use case driven, architecture-centric unvaried ,and progressive. the quantity of UML Diagram is offered.

V. ALGORITHM: CONVOLUTION NEURAL NETWORK (CNN)

1. Step 1: Input is given as image / video.
2. Step 2: Also numerous different pollutants are applied to the input to produce a point chart.
3. Step 3: Next a ReLU function is applied to increase non-linearity.
4. Step 4: Also applies a pooling subcaste to each and every point chart. Step 5: The algorithm compresses the pooled images into one long vector.
5. Step 6: In coming step, inputs the vector to the algorithm into a completely connected artificial neural network.
6. Step 7: Processes the features via the network. At the end completely connected subcaste delivers the “voting ” of the classes.
7. Step 8: In this last step trains through forward propagation and back propagation for multitudinous ages. This reiteration occurs until we've a well- defined neural network with trained weights and point sensors.

VI. . SCOPE

With the help of CNN algorithms, we can increase the rate of detection suspicious activity in the real-time Content. We can use it in the Banking sector, Home, IT sector, etc.

Conclusion

A system to method period CCTV footage to notice any suspicious activity can facilitate to form higher security and fewer human intervention. nice strides are created within the field of human suspicious Activity, that permits United States of America to raised serve the myriad applications that square measure potential with it. Moreover, analysis in connected fields like Activity following will greatly enhance its productive utilization in many fields.

References

[1] Hasan, M., Choi, J., Neumann, J., Roy-Chowdhury, A.K., Davis, L.S.: Learning temporal regularity in video sequences. In: 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). pp. 733–742 (June 2016) [2] C. Schuler, H. Burger, S. Harmeling, and B. Scholkopf. A machine learning approach for non-blind image deconvolution. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 1067– 1074, 2013 [3] W. Sultani, C. Chen and M. Shah, ”Real-World Anomaly Detection in Surveillance Videos,” 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, 2018, pp. 6479- 6488 [4] Kamthe, U. M., Patil, C. G. (2018) ”Suspicious Activity Recognition in Video Surveillance System” 2018 Fourth International Conference on Computing Communication Control and Automation (ICCUBEA). [5] Liu, C., Tao, Y., Liang, J., Li, K., Chen, Y. (2018) ”Object Detection Based on YOLO Network” 2018 IEEE 4th Information Technology and Mechatronics Engineering Conference (ITOEC) [6] Rohit Kumar Tiwari and Gyanendra K. Verma, “A Computer Vision based Framework for Visual Gun Detection using Harris Interest Point Detector”, Procedia Computer Science, vol 54, p. 703 - 712, 2015 [7] P.Bhagya Divya, S.Shalini, R.Deepa, Baddeli Sravya Reddy,“ Inspection of suspicious human activity in the crowd sourced areas captured in surveillance cameras”, International Research Journal of Engineering and Technology (IRJET), December 2017

Copyright

Copyright © 2023 Prof. Dipali Pawar, Omkar Dhanwat, Sushant Shrivastav, Devendra Sutar, Sourabh Yadav. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.