An Anomaly Detection Through Felis Territorial Search Optimized Continual Learning with Variational Auto-Encoder-Based Distributed Convolutional Neural Network (FeTSO-CnVDCN)

The advancements in surveillance cameras and their widespread deployment have shifted the research toward autonomous detection of suspicious activities. Anomaly detection in real time is an extremely challenging process and several approaches have been deployed for the cause. The methods show promising results however exhibit complexities associated with lower accuracy, inability to differentiate similar types of events, and inadaptability to pixel variations as well as detection in wide parameter settings. The research aims to propose Felis Territorial Search Optimized Continual learning with Variational auto-encoder-based Distributed Convolutional Neural network (FeTSO-CnVDCN) for addressing the limitations of the underlying approaches and to exhibit effective detection. The inclusion of a modified region-based convolutional neural network (MRCN) increases the accuracy by localizing the target with a bounding box. The Blockwise Local Arc Flowmap (BLArF) method extracts the color distributions, pixel orientations as well textures that provide better representations of the objects in the video frame. Moreover, the Felis Territorial Search optimization (FeTSO) algorithm tunes the weights of FeTSO-CnVDCN which improves the scalability towards diverse scenarios. The experimentation is performed against the conventional approaches and the proposed model achieves higher values of 97.49%accuracy, 0.97 Negative predictive value, and 0.98 Positive predictive value in the Shanghai tech database.