智能手表手势识别算法的研究

发布时间：2018-05-15 06:40

本文选题：智能手表 + 手势识别　；参考：《北京邮电大学》2016年硕士论文

【摘要】：近年来随着可穿戴设备的普及,智能手表这一新兴产品迅速抢占了市场并持续受到大众的广泛关注。作为便携和易用的象征,人们对智能手表的交互体验提出了新的要求。手势具有简单、高效、不受环境影响的特点,是最适合智能手表的交互方式。由于目前手势识别在智能手表上的应用程度还很有限,大多数都是使用基于固有规则匹配的识别方式。本文尝试对于成熟智能手表产品提出通用手势识别解决方案,并通过实验验证其有效性,从而为手表的功能扩展提供可行性依据。本文首先研究了当前常用的手势识别技术,对比其优劣之后最终确定了基于加速度的识别解决方案。根据实际使用需求对时间序列数据建模,并依据平台限制采用了基于动态时间规整和隐马尔可夫模型的算法框架。随后,考虑到手势数据的特殊性有针对地设计相应的辅助算法,并提出了基于最长连续子序列和回溯法的数据截取算法,有效地完成了数据的处理。本文随后按照上面提出的算法模型,基于Ticwear操作系统实现了智能手表端的手势识别系统。最后,结合研究需求设计了相应的实验,并使用该手势识别系统对假设模型进行了全面的验证和评测。结果表明,两种模型均能在300毫秒延迟之内给出预测结果,并且达到平均80%以上的识别正确率。从而证明了解决方案的可行性。
[Abstract]:With the popularity of wearable devices in recent years, smart watches, such a new product, quickly seize the market and continue to receive widespread public attention. As a portable and easy to use symbol, people have put forward new requirements for the interactive experience of smart watches. Gestures are simple, efficient, and are not affected by the environment. They are the most suitable for smart watches. Interactive mode. Because of the limited application of gesture recognition on smart watches, most of them use the recognition method based on the inherent rules matching. This paper tries to put forward a general gesture recognition solution for the mature intelligent watch products, and verifies its effectiveness by experiments, thus providing a feasible extension of the watch's function. First, this paper studies the current common gesture recognition technology. After comparing its advantages and disadvantages, the recognition solution based on acceleration is finally determined. The time series data is modeled according to the actual use requirement, and the algorithm framework based on dynamic time warping and hidden Markov model is adopted according to the platform constraints. Then, consideration is given. According to the particularity of the gesture data, the corresponding auxiliary algorithms are designed and the data interception algorithm based on the longest continuous subsequence and backtracking method is proposed, and the data processing is completed effectively. Then, based on the algorithm model proposed above, the gesture recognition system of the intelligent watch end is realized based on the Ticwear operating system. Finally, According to the research needs, the corresponding experiments are designed and the gesture recognition system is used to verify and evaluate the hypothesis. The results show that the two models can give the prediction results within 300 millisecond delay and reach an average of more than 80% recognition accuracy, which proves the feasibility of the solution.

【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41;TP368.33

【相似文献】