基于超声波信号的行为识别方法
Behavior recognition method based on ultrasonic signals
杨飏 1张雪2
作者信息
- 1. 中国海洋大学信息科学与工程学部,山东 青岛 266404
- 2. 武汉大学信息管理学院
- 折叠
摘要
作为近年来计算机视觉领域的研究热点,国内的行为识别技术主要依靠于不同场合的摄像头拍摄视频和图像进行分类识别、目标检测等相关视觉处理.因此,为了减少行为识别的消耗以及保护用户隐私,提出并实现了一种基于多普勒效应的行为识别技术,主要技术过程包括生成超声波信号、音频转换与降噪、连续音频切割、进行傅里叶变换以及卷积神经网络分类识别,行为识别的准确率可以达到 96.32%并且具备一定的泛化性能,应用前景广泛.
Abstract
As a research hotspot in the field of computer vision in recent years,the behavior recognition technology used in China mainly relies on the video and images captured by cameras for relevant visual processing such as classification,recognition and target detection.In order to reduce the consumption of behavior recognition and protect user privacy,a behavior recognition technology based on Doppler effect is proposed and implemented.The main technical process includes generating ultrasonic signal,audio conversion,noise reduction,continuous audio cutting,Fourier transform,and CNN classification and recognition.The accuracy of behavior recognition can reach 96.32%.It has certain generalization performance and broad application prospects.
关键词
超声波/多普勒效应/卷积神经网络/行为识别Key words
ultrasonic wave/Doppler effect/convolutional neural network(CNN)/behavior recognition引用本文复制引用
出版年
2023
NETL
NSTL
维普
万方数据