基于HOG算子的手型特征提取算法

HAND-SHAPE FEATURE EXTRACTION ALGORITHM BASED ON HOG

手型是动态手势的一个重要特征。已有的动态手势识别方法虽利用了手型信息,但是信息都隐含在数据中。对此提出一种动态手势中手型特征的提取算法。该算法能准确识别出动态手势中的手型特征。通过Kinect获取手部位置数据,计算手势样本的手局部图像的HOG(Histogram of Oriented Gradient)算子,通过统计手势过程中,临近帧的手局部图像的HOG算子的相似性,确定该样本的手型特征所在帧。实验结果显示,其手型特征识别平均正确率达到92.78%。

Hand-shape is an important feature of dynamic gestures. Though existing dynamic gesture recognition methods make use of the hand-type information,the information is hided in the data. We propose a new extraction algorithm for hand-shape feature in dynamic gestures,which can accurately recognise the hand-shape feature in dynamic gestures. It gets hand position data with Kinect,and calculates the HOG operator of partial hand image of gestures sample. The frame where the hand-shape feature of the gestures sample is found can be determined through the similarity of the HOG operators of the partial hand image in near frames in the process of counting the hand-shapes.Experiments show that average correct rate of hand-shape feature recognition achieves 92. 78%.