基于Haar小波分解的实时手势识别

传统基于样例的识别方法由于数据量大而难以应用于实时手势识别。为此，利用Haar小波可用于分解图像，保留细节部分而丢弃高频部分的特点，将视频序列中每一帧的图像尺寸降到不影响系统准确率的程度，同时降低识别过程的计算量。采用单摄像头在一个36个元素的美国手语手势集上进行实验，结果证明系统的有效识别速率可以提高到30 f/s且准确率几乎未变，可以满足实时性需求。

Real-time Gesture Recognition Based on Haar Wavelet Decomposition

Toward the basic exemplar-based recognition method is hard to apply in real-time gesture recognition for the huge amount of data, the Haar wavelet can be used to decompose images to keep the detail parts and discard the high-frequency parts, where each frame image in the video sequence can be downsized to the minimal level such that no obvious effect on the system＇s accuracy and the computational cost in the recognition stage can be saved. Comparing by the experiment with a single Web camera on a 36 elements of American sign language gesture dataset, the results show that the effective system recognition can be sped up to 30 f/s with the same accuracy, which meets real-time requirement.