基于头部跟踪的3D虚拟声源定位算法与应用  

3D virtual sound source localization algorithmand application based on head tracking

作　　者：张凤[1] 王发斌 夏珞珈 ZHANG Feng;WANG Fa-bin;XIA Luo-jia(College of Geographical Science and Tourism,Jilin Normal University,Siping 136000,China;College of Mathematics and Computer,Jilin Normal University,Siping 136000,China;College of Computer Science and Technology,Changchun University of Science and Technology,Changchun 130000,China)

机构地区：[1]吉林师范大学地理科学与旅游学院,吉林四平136000 [2]吉林师范大学数学与计算机学院,吉林四平136000 [3]长春理工大学计算机科学技术学院,吉林长春130000

出　　处：《吉林师范大学学报（自然科学版）》2024年第4期118-123,共6页Journal of Jilin Normal University:Natural Science Edition

基　　金：国家重点研发计划项目(2019YFC0409105);四平市科技发展计划项目(2023083)。

摘　　要：在头相关传输函数的基础上实现了基于头部跟踪的3D虚拟声源定位算法与应用,当使用者头部位置发生变化时,能够保持虚拟声源在空间中的绝对位置不变.本文通过Faster R-CNN目标检测算法,检测了摄像头画面中的人脸并实时返回人脸位置.根据返回的数值计算出水平角与俯仰角,使用CIPIC HRTF数据库中对应角度的测量值与输入声音信号进行卷积滤波处理,得到虚拟声音信号.实验表明,在向左向右分别偏移15cm和30cm的情况下,计算角度与实际角度的误差分别为0.36°、1.68°、0.41°、0.40°.Based on the head-related transfer function,a 3D virtual sound source localization algorithm and application based on head tracking have been implemented,which can maintain the absolute position of the virtual sound source in space unchanged when the user s head position changes.In this paper,the Faster R-CNN object detection algorithm was used to detect faces in the camera image and return the face position in real time.Based on the returned values,the horizontal and pitch angles were calculated.The measured values in the CIPIC HRTF database corresponding to the angle were convolved and filtered with the input sound signal to obtain the virtual sound signal.Experiments showed that when the head was shifted 15cm and 30cm to the left and right,respectively,the errors between the calculated angles and the actual angles were 0.36°,1.68°,0.41°,and 0.40°.

关 键 词：头部跟踪 虚拟声源 头相关传输函数 

分 类 号：F311.1[经济管理—产业经济]