融合时序注意力的行为识别

Action Recognition Based on Temporal Attention Fusion

为了获取图像序列中的时序信息,构建了卷积神经网络和时序注意力结合的网络模型。卷积神经网络用于捕捉空间特性,时序注意力机制用于捕捉序列间各样本之间的局部关联,通过对局部关联性加权求和得到序列时序特征。最后利用训练好的网络模型进行了测试,并对测试结果进行了展示。

In order to obtain the temporal information in the image sequence,a network model combining convolutional neural network and temporal attention is constructed.The convolutional neural network is used to capture the spatial characteristics,and the temporal attention mechanism is used to capture the local correlation between the samples in the sequence.The temporal characteristics of the sequence are obtained by weighted summation of the local correlation.Finally,the trained network model is used to test and the test results are shown.