Taylor series based compressive approach and Firefly support vector neural network for tracking and anomaly detection in crowded videos

الملخص EN

The application areas of multimedia content and computer vision analysis gains remarkable attention towards the motive to recognize the actions of the humans present in the video.

Accordingly, crowd behavior analysis is an important topic owing to the significance of video surveillance in the public localities.

This work introduces an anomaly detection (AD) model by introducing a tracking model and the optimization based classifier for the crowd video.

The objects present in the video require tracking since the anomaly depends on the action of the object.

This work proposes a hybrid tracking (HT) model by combining the Taylor series based predictive (TSP) tracking and the compressive tracking (CT) approach.

The features are extracted from the tracked objects, and a feature vector is formed.

Moreover, this work proposes the Firefly based support vector neural network (FSVNN) for the classification purpose.

The weights of the proposed FSVNN classifier are trained with the genetic and the firefly algorithm.

The performance of the proposed model is analyzed using three videos from the standard databases and is compared with the existing methods, such as self-organizing map (SOM), deep belief network (DBN), neural network (NN), and Firefly-based deep belief network (FDBN).

The simulation results reveal that the proposed AD model with the FSVNN classifier attains overall better performance with the values of 0.97035, 1, and 0.96, for sensitivity, specificity, and accuracy, respectively than the comparative methods.