[论文解读] Sequential Weakly Labeled Multi-Activity Recognition and Location on Wearable Sensors using Recurrent Attention Network

Kun Wang, Jun He+1 more|arXiv (Cornell University)|Apr 13, 2020

Context-Aware Activity Recognition Systems参考文献 23被引用 3

一句话总结

本文提出一种用于可穿戴传感器上顺序弱标签多活动识别与定位的循环注意力网络，支持在单个片段内对多个活动进行迭代注意力聚焦。该模型提升了准确率并实现了数据自动分割，在UniMiB-SHAR数据集和自定义顺序弱标签多活动数据集上优于CNN基线模型。

ABSTRACT

With the popularity and development of the wearable devices such as smartphones, human activity recognition (HAR) based on sensors has become as a key research area in human computer interaction and ubiquitous computing. The emergence of deep learning leads to a recent shift in the research of HAR, which requires massive strictly labeled data. In comparison with video data, activity data recorded from an accelerometer or gyroscope is often more difficult to interpret and segment. Recently, several attention mechanisms are proposed to handle the weakly labeled human activity data, which do not require accurate data annotation. However, these attention-based models can only handle the weakly labeled dataset whose segment includes one labeled activity, as a result it limits efficiency and practicality. In the paper, we proposed a recurrent attention network to handle sequential activity recognition and location tasks. The model can repeatedly perform steps of attention on multiple activities of one segment and each step is corresponding to the current focused activity according to its previous observations. The effectiveness of the recurrent attention model is validated by comparing with a baseline CNN, on the UniMiB-SHAR dataset and a collected sequential weakly labeled multi-activity dataset. The experiment results show that our recurrent attention model not only can perform single activity recognition tasks, but also can recognize and locate sequential weakly labeled multi-activity data. Besides, the recurrent attention can greatly facilitate the process of sensor data accumulation by automatically segmenting the regions of interest.

研究动机与目标

• 解决现有注意力模型仅能处理弱标签人类活动识别中单个活动片段的局限性。
• 实现在单个未标记片段内对多个顺序活动的准确识别与定位。
• 通过自动识别传感器数据中的感兴趣区域，减少对精确人工标注的依赖。
• 提升真实可穿戴传感器应用中的实用性与效率。

提出的方法

• 提出一种循环注意力机制，基于先前观察结果，在单个片段内迭代聚焦于不同活动。
• 使用可在多步中更新关注点的注意力模块，实现单个片段内多个活动的检测。
• 利用循环处理在注意力步骤间保持上下文，提升对顺序活动的识别能力。
• 将模型应用于基于加速度计和陀螺仪数据的活动识别与定位任务。
• 在仅提供片段级标签的弱标签数据上端到端训练网络。
• 采用双头架构，联合预测活动类别及其在片段内的时序位置。

实验结果

研究问题

• RQ1 循环注意力机制是否能有效识别单个弱标签片段内的多个顺序活动？
• RQ2 与标准CNN相比，循环注意力模型在弱标签多活动识别任务中的准确率表现如何？
• RQ3 该模型在无需精确标注的情况下，能在多大程度上自动分割原始传感器数据中的感兴趣区域？
• RQ4 在真实可穿戴传感器数据中，面对复杂且重叠的活动序列时，该模型是否仍能保持性能？

主要发现

• 与CNN基线相比，循环注意力模型在UniMiB-SHAR数据集和自定义顺序弱标签多活动数据集上均表现出更优性能。
• 该模型成功识别并定位了单个未标记片段内的多个顺序活动，克服了先前注意力模型仅支持单个活动的局限性。
• 循环注意力机制实现了感兴趣区域的自动分割，显著减少了人工数据整理工作量。
• 即使在活动序列复杂且重叠的传感器数据中，该模型仍保持高准确率。
• 注意力机制能动态在不同活动间转移关注点，基于先前上下文提升对后续动作的检测能力。
• 该方法在实际部署中展现出显著优势，大幅降低了对精确标注的依赖。