Anomaly Detection in Surveillance Videos Using Deep Learning and SVM Based Data Reduction Method
B K Hareesh Narayan, Abishi Chowdhury, Pankaj Shukla · 2024
Anomaly detection in video data is a crucial subject that has gained a significant research implication in computer vision with several real-time applications. The researchers worldwide developed innovative solutions in this context in the diverse areas of research specific practical applications. This paper emphasizes over the importance of anomaly detection in a public environment along with its applications for the same. A brief survey of prominent existing techniques with a comparative analysis has been presented in this paper in order to highlight the relevant techniques, available datasets, considered metrics for analysis, and future scope of improvements. In this context, deep learning (DL) approaches like Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks are prominently used for anomaly detection. In this paper, Support Vector Machine (SVM) along with convolution 2D LSTM based data reduction approach is presented which is used to keep only the important data over incremental addition of the data. The experimental analysis, performed on benchmark dataset DCSASS, shows the efficacy of the proposed approach when compared with state-of-the-art approaches as it maintains significant accuracy while ensuring effective data reduction. Furthermore, the proposed approach demonstrates the training and validation loss analysis. It reduces the errors as time progresses which is directly proportional to the incremental addition of data.