基于android平台的视觉手势识别分析-交通信息工程及控制专业论文.docxVIP

摘要 本文研究基于 Android 平台的视觉手势检测、手势分割和识别技术，实现了 在 Android 平台上对规定的六种静态手势的识别，为 Android 平台的人机交互应用 提供了新的技术手段。 首先，本文采用基于 LBP 特征的 Viola-Jones 目标检测算法实现手势检测，该 算法在人脸识别上已有广泛的应用，但在手势检测上还少有应用。该算法与基于 Haar-like 特征的 Viola-Jones 检测算法相比，在具有相当的检测率的同时，极大地 减少了计算量小，提高了实时性，因而适合在移动设备上应用。 其次，本文通过肤色分割、边缘检测、轮廓提取三个步骤实现手势分割，然 后提取分割后轮廓的傅里叶描述子，并使用主元分析对其进行降维处理，最后采 用 K 均值聚类的手势识别算法，对处理后的特征向量进行聚类分析，得到识别结 果。 最后，本文在搭载单核 1GHz 处理器、运行 Android 2.2 操作系统的华为 S7 Slim 平板电脑上，利用 Android NDK 编译工具和 OpenCV for Android 视觉算法库实现 了上述算法，并设计了 Android 手势识别应用编程接口，该接口目前可以对六种静 态手势进行检测和识别，最后调用该接口实现了一个使用特定手势控制摄像头完 成拍照的应用案例，其检测和识别一幅 352*288 像素视频图像帧的平均总时间为 91ms。 关键词：Android 手势检测 肤色分割 手势识别 K 均值聚类 Abstract This paper studies vision based gesture recognition related fields on Android system, including gesture detection, gesture segmentation and gesture recognition, designs and implements a static gesture recognition system on Android, this system can recognize 6 specific static gestures in real-time. It offers new interactive means for Android applications. First，for gesture detection, this paper presents the Viola-Jones gesture detection method based on LBP feature, which has been widely used in face recognition, but is found much less applications to gesture recognition. Compared to the Viola-Jones method based on Haar-like feature, this method has equivalent hit rate, but much less computations. So it is a better choice for embedded devices. Secondly, this paper introduces the hand segmentation scheme which consists of three processes: skin color detection, edge detection and contour extraction. Then the gesture recognition algorithm is presented, the 6 cluster centers of the 6 hand gesture is calculated by using PCA and K-means clustering, the recognition result is determined by the distance between a hand gesture feature vector and the 6 cluster centers. Finally, with the help of Android NDK compilation tools, and OpenCV for Android vision algorithm library，this paper implements the algorithm described above on Huawei S7 Sl