基于深度信息和显著计算的手势交互技术研究及应用

项目名称： 基于深度信息和显著计算的手势交互技术研究及应用

项目编号： No.61462038

项目类型： 地区科学基金项目

立项/批准年度： 2015

项目学科： 其他

项目作者： 杨文姬

作者单位： 江西农业大学

项目金额： 43万元

中文摘要： 手势交互技术是实现智能人机交互不可缺少的一项关键技术。基于视觉的手势交互技术易受光照、复杂背景和环境等影响，大大限制了它的实际应用。本项目拟对手势检测与分割、特征提取和手势建模与识别等手势交互技术的关键问题进行研究，以提高其适用性。基于现有手势检测技术易受光照和复杂背景影响，拟采用基于深度信息和显著计算相结合的方法进行静/动态手势检测与分割，大大提高手势检测的准确性。现有特征提取方法提取的大都是视角变化的特征，本项目拟在融合多种特征基础上构建一种视角稳定的特征。针对现有HMM观测建模方法需要估计大量模型参数，本项目拟采用核稀疏模型建模HMM观测，一方面可提高模型的表述能力，另一方面不会快速增加参数数目。在此基础上实现基于视觉的服务机器人手势交互系统。本项目的研究不仅能促进相关理论发展，而且可用于人与服务机器人间的交互，使得服务机器人能更好地融入人类社会，具有重要的科学价值和实际意义。

中文关键词： 深度信息；显著计算；手势交互；核稀疏隐马尔科夫模型；服务机器人

英文摘要： Gesture-based interaction is an important and indispensable technology for intelligent human-computer interaction. Vision-based gesture interaction technique is vulnerable to light, complex background and environment, which greatly limits its practical application. On the basis of the long-term research on the vision-based hand gesture interaction,The project will make research for key issues of gesture interaction technique to improve their applicability, namely gesture detection and segmentation, feature extraction and gesture modeling & recognition. Because the existing gesture detection method is susceptible to illumination changes and complex background, we plan to combine depth information with saliency computation to detect and segment hand in the static/dynamic gesture, which can achieve hand more accurately. Mostly existing feature extraction methods explore simple appearance feature for hand gesture recognition, therefore, the project intends to integrate a variety of features to build a stable and view-invariant feature. Due to a large number of model parameters to be estimated for existing observation models of HMM, the project intends to use kernel sparse representation to model observations of HMM, which can increase the expressive ability of the observation model and does not grow rapidly the number of parameters. Based on the above approaches, we will build and implement a vision-based gesture interaction system for service robot. The implementation of the project will not only promote the development of related theory , but also for the interaction between people and service robots ,which make service robots be integrated into human society better. Therefore, it has important scientific value and practical significance.

英文关键词： depth information;saliency computation;hand gesture interaction;kernel sparse HMM;server robot