An Ensemble Learning and Swarm Based Segmentation Framework for Video Event Recognition

Video event recognition aims to recognize the spatiotemporal visual patterns of events from videos. In recent years, event recognition has attracted growing interest from both academia and industry. Recognizing events in surveillance videos is still quite challenging, largely due to the tremendous intra class variations of events caused by visual appearance differences, target motion variations, viewpoint change and temporal variability. The existing system designed an extreme learning machine and action recognition algorithm for generalized maximum clique problem in video event recognition. The implemented system designed an enhanced ensemble deep learning and swarm based segmentation framework for video event recognition. The presented ensemble framework in that not only decreases the information loss and overfitting problems caused by single models. Initially, a video frames are taken as an input and most salient information extract from it. The VLAD for feature encoding is utilized for feature encoding. The segmentation process is done with the help of Random Inertia Weight based Particle Swarm Optimization (RIWPSO) of successive frames are exploited for pattern matching in a simple feature space. Thereafter, an Ensemble Learning (EL) is developed based on the performance of each SVM and Elman Recurrent Neural Network (ERNN) classifier on each feature set. Thus the simulation results demonstrate the effectiveness of the implemented enhanced ensemble deep learning technique for video event recognition compare to the existing methods.