摘要
针对3D舞美虚拟人物影像易受到成像条件干扰,导致人物姿态识别误差大的问题,设计基于激光视觉的3D舞美虚拟人物影像自动化识别系统。使用线结构光视觉三维测量装置,采集表演人物的激光视觉图像后,以平行投影的方式,重构3D舞美虚拟人物影像;通过基于谱系聚类的关键姿态轮廓特征提取方法,提取影像中关键姿态轮廓特征,输入基于HCRF的3D舞美虚拟人物影像人体姿态识别模型,计算人体姿态动作类型的条件概率,实现3D舞美虚拟人物影像自动化识别。实验结果表明,该系统对3D舞美虚拟人物影像中,行走、跑步、跳跃、跌倒、身体倾斜五种姿态类型自动化识别后,识别结果的交并比最小值为0.986,识别结果与真实姿态动作类型匹配。
In order to solve the problem that the 3D virtual figure image is susceptible to the interference of imaging conditions,which leads to the large error of character pose recognition,an automatic recognition system of 3D virtual figure image based on laser vision is designed.Using linear structured light vision 3D measuring device,the laser vision image of the performer is collected,and the 3D virtual figure image is reconstructed by parallel projection.Through the key pose contour feature extraction method based on lineage clustering,the key pose contour features in the image are extracted,and the human pose recognition model based on HCRF is input to calculate the conditional probability of the human pose action type,and the automatic recognition of 3D stage virtual figure image is realized.The experimental results show that after the automatic recognition of the five pose types of walking,running,jumping,falling and body tilt in the 3D dance beauty virtual figure image,the minimum crossover ratio of the recognition result is 0.986,and the recognition result matches the real pose type.
作者
胡巍
王英杰
HU Wei;WANG Yingjie(Academy of Military Culture,National Defence University of PLA,Beijing 100081,China;Space Engineering University,Beijing 101416,China)
出处
《电子设计工程》
2024年第14期180-184,共5页
Electronic Design Engineering
关键词
激光视觉
3D舞美
虚拟人物影像
姿态轮廓特征
姿态识别
laser vision
3D dance beauty
virtual character images
posture contour features
posture recognition
