基于改进的SOM和ART的动态手势识别

摘要

提出一种动态手势识别算法,将动态手势识别问题转换为轨迹识别间题.首先以SOM算法作为分类器提取手势特征,将多维手势信息投影到二维平面中,根据每帧的顺序产生一平面轨迹,将产生的平面轨迹输人到改进的ART网络进行识别.实验结果表明,该算法用于动态手势识别是可行的且性能稳定.%A dynamic gesture recognition algorithm was presented, which could convert the dynamic gesture recognition problem to track identification problem, First, the SOM algorithm as classifier was used to extract gesture characteristics, then the multi-dimensional gesture information was projected in the two-dimension plane. A plane trajectory was generated by the order of each frame, and the produced plane trajectory was input to identify by the improved ART network. The experiments showed that the algorithm for dynamic gestures was feasible, and its perforrnance was stable.