摘要
针对已有方法没有在识别的过程中对手势深度图像进行去噪处理,导致识别结果不准确,识别耗时较长的问题,提出一种基于全链条AI技术的手势深度交互识别方法。首先对含有噪声的手势图像进行轮廓变换,获取高频子带系数,根据阈值建立二元表,同时采用空间规则确定支持向量机的特征向量,获取去噪手势图像。然后采用自适应深度直方图阈值方法分别对手势深度图像和彩色图像进行分割,完成手势区域提取和肤色信息检测。最后结合全链条AI技术,对手势区域和肤色信息进行边缘匹配,获取准确的手势轮廓信息,实现手势深度交互识别。与已有方法进行测试对比,实验结果表明,所提方法能够快速、准确地完成手势深度交互识别。
In most approaches,the noise is not removed from deep gesture images during the recognition,so the recognition results are inaccurate and time-consuming.For this reason,a method of deep interactive recognition of gestures based on the whole-chain AI technology was proposed.Firstly,the contour transformation was performed on noisy gesture images,so that the high-frequency subband coefficients can be obtained.Secondly,the binary table was constructed according to threshold values.In the meanwhile,the spatial rule was adopted to determine the feature vectors of support vector machine and thus to obtain the denoising gesture image.Then,the adaptive deep histogram thresholding method was used to segment the deep image and the color image of gesture respectively,thus completing the extraction of gesture region and the detection of skin color information.Combined with the whole-chain AI technology,the edges of gesture region and skin color information were matched to obtain accurate gesture contour information.Finally,the deep interactive recognition of gestures was achieved.Compared with the existing methods,the new method can quickly and accurately complete the deep interaction recognition of gestures.
作者
张明星
陈彦卿
ZHANG Ming-xing;CHEN Yan-qing(Hunan University of Science and Engineering,Hunan Yongzhou 425199,China;School of Computer and Control Engineering,Qiqihar University,Heilongjiang Qiqihar 161003,China)
出处
《计算机仿真》
北大核心
2022年第7期252-256,共5页
Computer Simulation
基金
2018年湖南省哲学社会科学基金重点项目(18ZDB013)。
关键词
手势深度交互识别
轮廓变换
去噪
The whole-chain AI technology
Deep interactive recognition of gestures
Contour transforma-tion
Denoising
