Please use this identifier to cite or link to this item:
http://hdl.handle.net/10603/364961
Title: | Design and Development of Algorithms for Anomaly Detection in Video |
Researcher: | Gautam K S |
Guide(s): | Senthil Kumar T and Shunmuga Velayutham C |
Keywords: | Engineering and Technology;Computer Science; Imaging Science and Photographic Technology; Deep learning; anomaly detection; soft computing; face recognition; Image Analysis |
University: | Amrita Vishwa Vidyapeetham University |
Completed Date: | 2021 |
Abstract: | Video surveillance is one of the important features that ensure the security of a building.Surveillance is made intelligent when the system can automatically detect the anomalies in the given scene. The research aims to identify and categorize anomalies. The datasets have been captured in Smart Spaces Lab funded by Department of Science and Technology (DST), India. The proposed anomaly detection algorithms monitor the area under surveillance and raise an alert in case of an abnormal event. The research suggests algorithms that detect anomalies such as unauthorized entry, possession of hidden weapon, and identifying negative emotions. To identify unauthorized entry a hybrid algorithm is recommended using skin detector and Haar cascade classifier (HCC) that could detect the face image across orientation changes and the detected face images are identified using Eigenfaces. Even when the input is a tilted face, the model detects and identifies the face. The algorithm proposed to detect the face image is Deep Learning based and built using Haar Cascade Classifier and Skin Detector (HCCSD). This algorithm attains a precision of 99.10%. The algorithm overcomes limitations such as illumination, partial occlusion and face orientation change and detects face image with higher precision. newlineThe second part of this research proposes a Modified K-means Segmentation algorithm (MKS) that unambiguously sections covered objects. Our investigation addresses this issue by providing a robust practical solution. The proposed algorithm assists to choose the optimal value of K to segment the objects. System validation is done with images from Stereo Thermal Dataset and the results achieve a precision of 88.89% for object segmentation. The experimental outcome confirms that the proposed algorithm is perfect to be used in object segmentation without losing its number and shape. The achieved performance in terms of Top 1 Accuracy is 0.94. The system has the scope extended to the areas like prisons, airports, and so on where the necessity... |
Pagination: | xii, 111 |
URI: | http://hdl.handle.net/10603/364961 |
Appears in Departments: | Department of Computer Science and Engineering (Amrita School of Engineering) |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
01_title.pdf | Attached File | 252.65 kB | Adobe PDF | View/Open |
02_certificate.pdf | 202.84 kB | Adobe PDF | View/Open | |
03_preliminary pages.pdf | 365.26 kB | Adobe PDF | View/Open | |
04_chapter 1.pdf | 414.11 kB | Adobe PDF | View/Open | |
05_chapter 2.pdf | 319.1 kB | Adobe PDF | View/Open | |
06_chapter 3.pdf | 937.05 kB | Adobe PDF | View/Open | |
07_chapter 4.pdf | 586.66 kB | Adobe PDF | View/Open | |
08_chapter 5.pdf | 702.65 kB | Adobe PDF | View/Open | |
09_chapter 6.pdf | 288.93 kB | Adobe PDF | View/Open | |
10_bibliography.pdf | 307.61 kB | Adobe PDF | View/Open | |
11_publications.pdf | 181.64 kB | Adobe PDF | View/Open | |
80_recommendation.pdf | 541.14 kB | Adobe PDF | View/Open |
Items in Shodhganga are licensed under Creative Commons Licence Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0).
Altmetric Badge: