摘要
人体行为识别技术广泛应用于工业安全系统、第三方支付系统等领域。提出的方法遵循K?püklü等人提出的YOWO架构,在其基础上,重新构建通道融合模块以及重写边界回归策略的部分算法:(1)在通道融合与注意力机制模块的基础上添加average pooling模块,进而达到提高待训练模型学习能力的效果;(2)重新设计ResNext-101模块,提高模型的表达能力,同时采用CIoU回归损失函数来衡量模型输出与真实标注框之间的差异对模型进行收敛,最终提高边界框回归的稳定性。在公开数据集UCF101-24和J-HMDB-21上的实验结果表明,该方法能够有效地增强视频特征的表达能力,在识别与定位人体行为视频的检测精度、定位精度以及稳定性上优于相应的同类算法。
Human behavior recognition technology is widely used in industrial security systems,third-party payment systems and other fields.The method proposed in this paper follows the yoyo architecture proposed by K?püklüet al.On its basis,it reconstructs the channel fusion module and rewrites some algorithms of boundary regression strategy:(1)Add the average pooling module method to the channel fusion and attention mechanism module,So as to improve the learning ability of the model to be trained.(2)Redesign the resnext-101 module to improve the expression ability of the model.At the same time,the CIO regression loss function is used to measure the difference between the model output and the real annotation box to converge the model,and finally improve the stability of the bounding box regression.The experimental results on the public data sets UCF101-24 and J-HMDB-21 show that the method proposed in this paper can effectively enhance the expression ability of video features.It is superior to the corresponding similar algorithms in the detection accuracy,location accuracy and stability of human behavior video recognition and location.
出处
《工业控制计算机》
2023年第1期100-101,104,共3页
Industrial Control Computer
关键词
YOWO
特征提取
相关系数矩阵
人体行为识别
损失函数
YOWO
feature extraction
corresponding coefficient matrix
human behavior recognition
loss function
