Towards Complex Wavelet and Gaussian Features for Human Action Recognition
In this paper, we have proposed an efficient feature extraction technique for recognizing human action from videos. The novelty of feature extraction technique is its combined form and this combination is accomplished by combining two different feature extraction techniques such as Wavelet feature sand Gaussian features. For a given input action video, this approach extracts both the Difference of Gaussian feature and complex wavelet features and fuses them to form a single feature vector. Further for classification, we have employed two classifiers such as K-Nearest Neighbor (K-NN), ad Support Vector Machine (SVM). Simulation experiments are conducted over a standard benchmark dataset, i.e., UCF YouTube Action dataset. The performance of developed HAR system is analyzed through Recognition Accuracy. Keywords - Action Recognition, Difference of Gaussian, Complex Wavelet, UCF action dataset, Accuracy.