摘要
在运动员动作图像特征提取的研究中,为了对运动员动作进行准确评估,利用动作图像特征进行区域性精细划分,获得不同层次的动作图像特征集合是特征提取的前提。传统反向传播增强算法只能将动作图像粗分为几个子类,不能获取准确的动作特征集合,导致出现动作图像特征提取效果差的问题。提出基于层级化特征的运动员动作图像特征提取方法。对运动员运动技术形成图像特征底层特征进行分析,对图像特征进行区域性的划分,得到不同层次形成动作的特征集合,并让图像特征映射到相对应的特征维度空间上,利用AdaBoost算法筛选出对智能视觉分析贡献最大运动员动作图像的特征数据,并作为训练样本进行训练与识别,完成对运动员动作图像特征提取。仿真结果证明,基于层级化特征的运动员动作图像特征提取方法可以对运动员动作图像特征进行准确提取。
For accurate assessing athlete action, we used the action image features to make regional partition subtly during the research on feature extraction of athlete action images. The precondition of features extraction to obtain the features set of action image in different levels. Traditional counter propagation enhancing algorithm can only divide the action image to some subclass roughly. It cannot obtain the accurate set, which leads to a poor effect of features extraction. In this paper, we proposed a feature extraction method of athlete action image based on the stratification feature. Firstly, the we analyzed the low-level feature of image features formed by athlete behavioral technology. Then, we made regional partition to image feature to obtain the feature set of action formed by different layers and mapped the feature to its corresponding feature dimension space. Moreover, we adopted the AdaBoost arithmetic to screen out the feature data of action image contributing to intelligent visual analysis most greatly and used it as the training sample to make training and recognition. Finally, we finished the feature extraction of athlete action image. The simulation results indicate that the method based on stratification features can accurately extract the action image feature.
出处
《计算机仿真》
北大核心
2017年第1期384-387,共4页
Computer Simulation
关键词
特征提取
智能视觉
层级化特征
Feature extraction
Intelligent vision
Stratification features