Comparative Analysis of Event Detection in Video Sequences using Adaptive Nature-Inspired PBCDNN, Ensemble FS-Linear MSVM and CNN Techniques

Nowadays, the computer vision (CV) society has defined a strong interest in the current growth of event detection and classification in UCF-101 dataset videos. UCF-101 dataset has different classes, such as basketball, pushups, sports, etc., and the increase in sports has expanded the technical focus of CV society, prompting various research. Videos and images remain the most commonly used elements in the CV–several methods and approaches can be used to classify the videos. This research article implements a new strategy and comparative analysis to detect various events in the video sequences. Firstly, it was gathered from UCF-101 and CCV datasets. Pre-processing steps enhance image quality, noise reduction, and frame extraction. After that, the feature extraction uses the HoG method to extract the feature sets that speed up the training procedure. A hybrid feature selection method is implemented to choose optimal features. The selected optimized features (adaptive nature-inspired PBC) are fed to the DNN classifier to classify several events. The proposed model has shown a maximum of 21.3% accuracy and reduced error rates in the VED and classification by accuracy compared to the existing methods, such as Ensemble FS-linear MSVM, CNN, K-nearest neighbour (KNN), and Random Forest (RF).