基于混合式协同训练的人体动作识别算法研究

摘要

Human action recognition is an important issue in the field of computer vision.Existing action recognition methods mostly belong to supervised learning category,in which a large number of labeled data are needed to train the recognition model.However,in many real-world tasks,labeled data are often expensive to get,while unlabeled data are readily available in abundance.In this paper,a novel human action recognition algorithm,named as Co-KNN-SVM,was proposed based on hybrid collaborative training.Different types of recognition methods for action recognition field are employed in this method to build the base classifiers,which are then iteratively retrained to increase their generalization abilities.In general,our method can decrease the labeling cost and achieve complementary advantages of different recognition algorithms.In order to decrease the impact of the noise in pseudo labeled data and improved the recognition performance,we also improved the selection method for the pseudo label data and the iterative training strategy in co-trai-ning style algorithms.The experimental results show that the proposed algorithms can identify human action in the vi-deo more effectively.%人体动作识别是计算机视觉研究中备受关注的课题.现有的动作识别方法大多属于监督学习,需要大量的有标记数据来训练识别模型.然而,在现实应用中有标记的数据成本较高,而无标记数据很容易获取.提出一种基于混合式协同训练的新型人体动作识别算法--Co-KNN-SVM,该算法利用动作识别领域不同类型的方法来构建基分类器,并进行迭代的相互训练以提高泛化性能,可以降低标注成本,并实现不同识别方法的优势互补.此外,还改进了协同训练中对伪标记数据的选择方法和迭代训练策略,有效控制了伪标记数据的噪声影响,提高了协同训练的识别效果.实验结果表明,所提算法可以有效地识别视频中的人体动作.