基于Kinect遮挡条件下行人的深度图像分割  

Depth image segment based on Kinect occlusion condition

作　　者：刘伟[1] 

机构地区：[1]重庆邮电大学通信与信息工程学院,重庆400065

出　　处：《重庆邮电大学学报（自然科学版）》2014年第2期271-275,共5页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

摘　　要：针对Kinect深度图像中有遮挡条件下的多个行人进行分割算法的实时性和应用研究,提出一种双峰法和迭代法的自适应阈值改进算法,并融合平面位置关系消除重复目标的新算法。在提取遮挡行人目标的连通域后,在连通域的遮挡条件下,对行人有不同深度数据的特性计算出多个阈值,快速、有效地分割行人,对分割后深度图像用改进的消除多余目标算法使分割深度图像的结果更为准确。实验结果表明,在使用该新阈值分割和消除重复目标算法后,行人量大和行人量小时处理一帧需要60 ms—70 ms,正确识别率为94.8%—97.3%。从结果得出,新深度图像分割算法具有更好的实时性和鲁棒性,能良好地应用于Kinect深度图像分割中。According to the real-time performance and application study of segmentation algorithm of occlusion pedestrians in a depth image, which is for Kinect depth image, we study and present an adaptive threshold algorithm based on two-peak method and iterative method. Under the circumstance of depth data segment, this new algorithm uses position relationship to eliminate repeated objectives . The algorithm calculates several threshold values from various depth data of occlusion pe- destrians which are connected, so as to segment the image of quickly and efficiently. Moreover using improved algorithms to eliminate redundant target after depth image segmentation makes the segment more accurate. It can finish processing one image during 60 ms - 70 ms whose correct recognition rate is during 94.8% - 97.3%. The results show that the new depth image segmentation algorithm has better real-time performance and robustness, which can be used well in Kinect depth im-age.

关 键 词：视频检测 目标合并 深度分割 行人遮挡 KINECT 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]