基于显著性区域的红外行为识别

视频特征的提取是行为识别方法中一个关键步骤，当视频场景中存在无关行人或者背景干扰时，提取的特征往往会包含较多的干扰信息，这将严重影响分类器的分类效果，进而影响行为识别准确率。针对这类问题，提出了一种基于显著性区域的红外行为识别方法。该方法对视频序列提取光流运动历史图(optical flow motion history image, OF-MHI)特征，获取视频序列的运动信息，此步骤旨在消除图像背景及静止目标干扰。利用类别激活映射(class activation map, CAM)方法进一步消除运动目标干扰，获得兴趣目标显著性区域，进而获得显著性区域特征图。输入卷积神经网络(convolutional neural network, CNN)提取最终特征，并采用支持向量机(support vector machine, SVM)获得识别结果。与传统方法相比，实验结果表明，该方法有效地提升了识别准确率。

Infrared action recognition method based on saliency region

The extraction of video features is a key step in action recognition methods. However, when unrelated pedestrians or backgrounds exist in the scene, the extracted features often contain more interference information, which leads to low recognition accuracy. To solve this problem, this paper proposes an infrared action recognition method based on saliency region. Firstly, the proposed method extracts optical flow motion history image (OF-MHI) features from the image sequences, obtaining the motion information of the videos to eliminate the interference of backgrounds and static targets. Then, we use the class activation map (CAM) method to eliminate the interference from moving targets to get saliency regions, thus obtaining significant regional features. Finally, the convolutional neural network (CNN) is used to extract the final features, and then we input the extracted features into the support vector machine (SVM) to obtain the recognition results. Compared with the traditional methods, the experimental results show that the proposed method can effectively improve the recognition accuracy.