Human Behavior Recognition with High-Order Attention Mechanism

Because the existing video behavior recognition methods had the problem that ignore the interactional relationship among features in the process of feature extraction, the effect of distinguishing approximate actions was poor. Therefore, a human behavior recognition method with high-order attention mechanism was proposed. A high-order attention module was introduced to a deep convolutional neural network, which modeled and utilized the complex and high-order statistics information in attention mechanism. The goal of attention was to reallocate the weight of each part of the feature map in the process of training, so as to focus on the local fine-grained information, produce the discriminative attention proposals, and capture the subtle differences among behaviors. Extensive experiments had been conducted to validate the superiority of our method for action recognition on two human behavior datasets, including UCF101 and HMDB51. The results showed that the recognition rate has been improved to a certain extent compare with the existing methods, which validated the effectiveness and robustness of the proposed method. The method effectively improves the ability to distinguish approximate behaviors.