On integration of multiple features for human activity recognition in video sequences

Human activity recognition has become one of the most active areas of research in computer vision, due to its increasing demand in many automated monitoring applications such as visual surveillance, human-computer interaction, health care, security systems, and many more. This work aims to introduce an integrated feature descriptor which combines texture feature and shape feature, at multiple orientations, to construct the efficient and robust feature vector for activity recognition in realistic scenarios. This feature descriptor is an integration of Discrete Wavelet Transform (DWT), multiscale Local Binary Pattern, and Histogram of Oriented Gradients (HOG). HOG descriptor extracts local-oriented histograms of the frame sequences, multiscale LBP gives the complex structural information of the frames and DWT gives the directional information at multiple scales. By exploiting these properties, we have constructed an integrated feature descriptor to construct the feature vector and achieves promising results of activity recognition in realistic videos. Multiclass Support Vector Machine (SVM) classifier with one-vs-one architecture has been used for activity recognition. The experiments are performed on five benchmark publicly available video datasets, namely Weizmann, IXMAS, UT Interaction, HMDB51, and UCF101. The experimental results are compared with the results of other state-of-art methods based on conventional machine learning and deep learning-based methods to show the effectiveness and usefulness of the proposed work. The experimental results have demonstrated that the proposed method performs better than the other state-of-art methods.