Behavioral Classification Using Feature Selection in the Micro Activity Retrieval Task
Behavioral Classification Using Feature Selection in the Micro Activity Retrieval Task
复制标题
DOI：
--
复制
发表时间：
--
期刊：
影响因子：
--
通讯作者：
Takuma Yoshimura;Pham HuuLong;Ryota Mibayashi;Rui Kimura;Hiroaki Ohshima
中科院分区：
文献类型：
--
作者：
Takuma Yoshimura;Pham HuuLong;Ryota Mibayashi;Rui Kimura;Hiroaki Ohshima
人类活动识别在帮助计算机理解人类活动方面具有重要作用。其中，微活动识别需要计算机理解更详细的动作。在这项研究中，我们提出了一个微活动检索任务的方法，以防止过度学习的数据集与少量的数据和许多特征维度。作为特征选择方法，我们使用Super-LCC。Super-LCC具有速度快、信息熵损失小的特点。该方法将特征尺寸从3108维平均减少到20.75维。所提出方法的mAP评价结果为0.71707。我们成功地完成了这项任务，采用了一种需要输入更少特征尺寸的技术。
Human activity recognition has an important role in helping computers to understand human activity. Among them, micro activity recognition is required for computers to understand more detailed action. In this research, we proposed a micro activity retrieval task method to prevent over-learning for data sets with a small number of data and many feature dimensions. As the feature selection method, we use Super-LCC. Super-LCC is fast and has low information entropy loss. The proposed method reduces the feature size from 3108 dimensions to 20.75 dimensions on average. The result of the mAP evaluation of the proposed method was 0.71707. We succeeded in the task with a technique that requires fewer feature dimensions to be entered.