基于近程电容传感器的3D交互技术研究

In the application of human-computer interaction, capacitive proximity sensor has the advantages of low-cost (it can be made by any conductive materials), design flexibility, high durability and stability, and it is insensitive to light and complex background. However, its short detection range and poor accuracy of 3D location limit its wide application. Therefore, we propose a combine method of operation model of determinant sensing configuration and BP feed-forward neural network approach. Two or more rows/columns of electrodes of the sensor array are used to sense the hand motion to increase the detection range of the sensor array because of the increased sensing area of the working electrodes. Then, the 14 determinant capacitive sensors are employed for hand detection at the same time. The measurement capacitances and the current coordinates are used as the input and output of the BP neural network approach to train the BP natural network model. The model with the optimized parameters, which is trained by the large sample of different hand shapes, is used to evaluate the position of the unknown hand. The approach with multi-input enables a high accuracy of 3D location. Thus, the contradiction between detection range and positioning accuracy is solved. At last, we use the sparse representation to recognize 3D natural interaction gesture. The proposed approach enables full spatial and rotation invariance and provides high recognition rate. In the project, we propose a whole method for interaction operation. The method based-on capacitive proximity sensors is very suitable for human-computer interaction applications in the future, such as exhibition hall, smart home, etc. Meanwhile, the 3D location approach can be widely used for 3D location and distance measurement.

近程电容传感器因成本低（导电体即可制作）、设计灵活、稳定性与耐用性好及不受光线与复杂背景影响，在人机交互应用中具有一定优势，但其探测距离近、三维定位精度差也制约了更广泛的应用。因此，我们提出一种行列式敏感配置操作模式与基于BP神经网络算法结合方法。该方法用两或多行/列电极去探测手部运动，通过增加工作电极面积提高探测距离；利用敏感配置的14个行列式传感器对同一时刻手部进行检测，将测量值及与之对应手部空间坐标作为基于BP神经网络的模型输入量和输出量。用不同手型实现建立的大样本训练优化的模型参数，用以评估未知手部的空间坐标，可获得高定位精度。这样解决了探测距离与定位精度的矛盾。最后，我们应用稀疏表示方法对3D自然交互手势识别，该方法可保持手势的空间与旋转不变性，提高手势识别率。本项目中提出一套完整的交互方法，可应用于展厅展示、智能家居等场景，同时，三维空间定位方法可适用广泛的定位与定距应用。

人机交互是近年来人工智能的热点之一，采用最方便和低成本的交互设备和形式更容易获得市场推广与应用。当下流行的基于机器视觉交互，存在成本高、数据量大、受光线与复杂背景影响及隐私泄露隐患等问题。近程电容传感器可以很好地解决以上问题，且具有更灵活的用户个性化设计与无死角安装等优点，在智能家居交互应用中具有显著优势，应用前景广阔。同时其探测距离近、三维定位精度差及交互方式单一等不足也制约了更广泛的应用。为解决了以上问题，本项目的主要研究工作如下：.1. 设计了高灵敏度的信号检测电路，提高了传感器的探测距离；.2. 提出了行列式敏感配置方式，并结合机器学习显著提升了电容传感器阵列的精确定距定位，解决了电极面积与分辨率的矛盾；.3. 实现了基于机器学习的静动态交互手势的识别；.4. 搭建了基于电容传感阵列的人机交互原理样机，验证了方案的实际可行性，为后续项目能真正应用落地提供了理论与实践基础。.在项目开展研究过程中，发表SCI期刊论文3篇，2篇在投，国内外会议论文3篇，授权专利1项，申请专利4项。本项目研究成果可用于智能家居、人体姿态监测等应用场景。

内容获取失败，请点击重试