应用交叉覆盖算法实现静态手势识别

REALISING STATIC GESTURE RECOGNITION WITH ALTERNATIVE COVERING ALGORITHM

静态手势识别是实现人机交互的前提和基础,多数手势识别采用模板匹配或人工神经网络算法。基于交叉覆盖算法的手势识别,首先通过摄像头采集到的手势图像经过灰度变换、平滑、二值化等预处理,在训练阶段,用交叉覆盖算法对二值化手势图像进行训练以得到手势分类,最后在测试集上进行手势识别。实验结果表明,由于该算法避免了优化过程中所需的巨大运算量,且允许分类目标在一定范围内的动态变化,手势识别的准确率得到有效提高。

Static hand gesture recognition is the prerequisite and base of the human-computer interaction. Most of the hand gesture recognition systems use template matching or artificial neural network algorithm, while ours in the paper is based on the alternative covering algorithm. First, we take photos of different hand gestures, and pre-process them with gray transforming, smoothing and binarising, etc. ; then in training stage, we use alternative covering algorithm to train these binarised hand gesture images to obtain the categorisation of hand gestures; at last, we reeognise these hand gestures on test set. Experimental result shows that since the alternative covering avoids huge computation during the process of optimisation, and also allows the dynamic variation of the classifying targets within certain scope, the accuracy rate of the hand gesture recognition is enhanced effectively.