为了提高基于加速度传感器的动态手势识别算法的性能,并且增强系统的可扩展性,提出了一种有效结合机器学习模型与模板匹配的方法.将手势分为基本手势和复杂手势两大类,其中复杂手势可分割为基本手势组成的序列;根据手势运动的特点提取有效的特征量,并利用基本手势样本训练随机森林模型,然后用其对基本手势序列进行分类预测;将预测结果进行约翰逊编码,再与标准模板序列进行相似度匹配.实验结果表明,该方法获得了99.75%的基本手势识别率以及100%的复杂手势识别率.算法既保证了手势识别的精度,也提高了系统的可扩展性.%This paper presents an algorithm combining machine learning model and template matching to improve the performance of accelerometer-based dynamic hand gesture recognition and enhance the extensibility of the system. Gestures are divided into two types,i.e.,the basic gesture and the complex gesture which can be decomposed into a basic gesture sequence. According to the characteristics of hand movements,effective features are extracted. A ran-dom forest model is constructed with the basic gesture samples,and then used to classify the basic gesture sequences. The predicted results are subsequently encoded with Johnson codes,and then matched with the standard template sequences by comparing the similarity. Experiment achieves 99.75% basic gesture recognition rate and 100%complex gesture recognition rate. The algorithm improves the performance as well as enhances the extensibility.
展开▼
