基于注意机制的仿生人体动作识别

针对已有的仿生动作识别系统计算量大、耗时长的问题,提出了在系统中加入注意机制的方法.该方法模仿注意机制在人类视觉中所起的重要作用,将视频中运动目标的活动区域作为视频图像的显著区域划分出来,然后结合视频图像的显著区域获取不同动作的特征模板.此方法在一定程度上克服了视频的背景噪声对系统性能的影响,同时更准确地模拟了人类视觉机制.实验结果表明：该方法能较好地改善人体动作识别模型的识别速度和识别结果.

A Biologically Inspired System Based on Attention Mechanism for Action Recognition

A method is introduced to improve the time-consuming problem of the already existed biologically-motivated system for recognition of action.Inspired by the importantance of attention mechanism in the human visual system,this method extracts the object’s movement area and regard it as picture’s saliency map,then the feature prototypes were extracted depend on saliency map.This method can overcome the adverse effect of background noise in the video,at the same time,it makes our system fit with human visual mechanism more.Experimental results show that our method can improve the model’s rate of identification and speed.