Sparse Representation for Crowd Attributes Recognition

Human behavior analysis has become a critical area of research in computer vision and artificial intelligence research community. In recent years, video surveillance systems of crowd scenes have witnessed an increased demand in different applications, such as safety, security, entertainment, and personal mental health. Although many methods have been proposed, certain limitations exist, and many unresolved issues remain open. In this paper, we proposed a novel spatio-temporal sparse coding representation, based on sparse coded features with k-means singular value decomposition for robust classification of crowd behaviors. Extensive experiments have shown that dictionary learning method with sparsely coded features captured vital structures of video scenes and yielded discriminant descriptors for classifications than conventional bag-of-visual-features. Relying on the measurable features of crowd scenes and motion characteristics, we can represent different attributes of the crowd scenes. Experiments on hundreds of video scenes were carried out on publicly available datasets. Quantitative evaluation indicates that the proposed model display superior accuracy, precision, and recall in classifying human behaviors with linear support vector machine when compared with the state-of-the-art methods. The proposed method is conceptually simple and easy to train: thereby achieving an accuracy of 93.50%, a precision of 93.40%, and a recall of 95.96%.