摘要
针对场景的光照变化和遮挡、混响等因素对目标定位准确性和鲁棒性的影响,提出了一种基于Kinect音视频融合的目标定位方法.在获取场景的颜色、深度和声源定位信息后,首先利用获取的深度信息剔除背景信息,然后分别对颜色、深度和声源定位的模型计算似然函数,最后融合上述3种似然函数,并在粒子滤波框架下实现目标定位.实验结果表明,音视频信息融合的目标定位平均准确率达到90.7%,相比于同一场景下独立使用视频和音频定位的准确率分别提高了9.1%和16.9%.
Aiming at the influence of illumination variation, occlusion and reverberation on the accuracy and robustness of object location, a new method based on Kinect audio-video fusion is proposed. After obtain the color, depth and sound source location information of the scene, firstly, the background information is eliminated by depth information. Secondly, the likelihood function is computed for the model of color, depth and sound source location. Finally, fused three likelihood functions and implemented target location under framework of the particle filter. The experimental results show that the average accuracy of proposed method reaches to 90.7%, in contrast with singly using video and sound source location methods, the proposed method increased accuracy by 9.1% and 16.9% respectively.
作者
李思含
罗凯
金小峰
LI Sihan;LUO Kai;JIN Xiaofeng(College of Engineering, Yanbian University, Yanji 133002, China)
出处
《延边大学学报（自然科学版）》
CAS
2018年第1期69-73,共5页
Journal of Yanbian University（Natural Science Edition）
基金
吉林省科技厅自然科学基金资助项目(20140101225JC)
关键词
KINECT
信息融合
深度信息
目标定位
声源定位
粒子滤波
Kinect
information fusion
depth information
target localization
sound source localization
particle filter
