欢迎访问《图学学报》
分享到：
摘 要：该文研究并提出了一种轨迹引导下的举重视频关键姿态自动提取方法。#br# 针对举重训练，首先提取稳定的杠铃轨迹，进一步分析杠铃轨迹和关键姿态之间的关系，将#br# 杠铃轨迹和基于姿态集的方法相结合进行关键姿态检测。根据运动轨迹的曲线极值点提取关#br# 键视频画面，而对于其他非轨迹极值点处的关键画面采用基于姿态集的姿态估计和目标检测#br# 方法，对每个关键姿态分别训练了一个线性的支持向量机分类器，建立图像的多尺度扫描模#br# 式，并提出了统计计算相似度的方法来处理帧间相似度问题，实验表明该文方法在姿态检测#br# 的准确性和效率方面都有很大改善。#br# 关 键 词：轨迹；姿态；关键帧；支持向量机
Abstract: In this paper, a trajectory guided scheme is proposed to extract the key poses#br# frame automatically. First, the barbell trajectory is extracted from the weight lifting sport video.#br# Then the barbell trajectory and poselet are combined for pose extraction. Some key poses are#br# extracted from the extreme point of the barbell trajectory. But the other poses are extracted by#br# using the poselet based algorithm. SVM(Support Vector Machine) classifiers based#br# HOG(Histogram of Gradient) is trained for each pose. Then the pose is detected in the multi-scale#br# images. Statistical similarity is computed in multi-scale to measure the possibility and determine#br# which is the best key frame. It resolves the problem of higher inter-frame similarity. The#br# experimental results show that the proposed scheme can improve both the precision and#br# performance of pose extraction.#br# Key words: trajectory; pose; key frame; support vector machine
摘要： 该文研究并提出了一种轨迹引导下的举重视频关键姿态自动提取方法。
针对举重训练，首先提取稳定的杠铃轨迹，进一步分析杠铃轨迹和关键姿态之间的关系，将
杠铃轨迹和基于姿态集的方法相结合进行关键姿态检测。根据运动轨迹的曲线极值点提取关
键视频画面，而对于其他非轨迹极值点处的关键画面采用基于姿态集的姿态估计和目标检测
方法，对每个关键姿态分别训练了一个线性的支持向量机分类器，建立图像的多尺度扫描模
式，并提出了统计计算相似度的方法来处理帧间相似度问题，实验表明该文方法在姿态检测
的准确性和效率方面都有很大改善。
