在基于几何模型的手势识别方法中，尺度空间特征检测是一种最常用的方法。由于传统方法涉及大量的高斯卷积运算，计算非常复杂。提出了一种快速的尺度空间特征检测方法，采用一组简单的矩形特征模板近似传统方法中复杂的高斯导数卷积模板，得到了尺度空间几何特征的快速检测子。通过对手势图像中Blob和Ridge结构的检测，得到手掌和手指结构的描述，进而完成手势识别。矩形特征模板的卷积可以用积分图进行快速计算，该方法使特征检测的速度得到了很大提高。在标准数据集和自然环境图像数据上的实验结果表明，该方法在保证识别准确率的同时，有效地提高了手势识别的实时性。

Scale-space feature detection is one of the most frequently used method in hand gesture recognition based on geometric model. However, the traditional method of scale-space feature detection involves heavy computation of Gaussian convolution, which makes the detection and recognition time-costly. In this paper, a fast scale-space feature detection method is proposed. First, a series of simple rectangular feature templates are used to approximate the complicated Gaussian derivatives convolution templates, with which the fast detectors of scale-space geometric features are obtained. After the detection of blob and ridge structures in gesture image, palm and finger structures are described and then gesture recognition is performed according to the configuration of palm and fingers. Then, integral image is used to rapidly calculate the convolution of rectangular feature templates, so the detection of scale-space geometric features is greatly accelerated in the method. Experiments on the standard dataset and the natural scene dataset show that the proposed method significantly reduces the time cost of gesture recognition while keeping comparable accuracy with traditional method.