一种基于特征增强和决策融合的人体行为识别方法

Method of Human Activity Recognition Based on Feature Enhancement and Decision Fusion

利用三轴加速度传感器进行人体行为识别一直是传感器数据处理、模式识别领域的研究热点。加速度数据往往存在着多种动作数据难以区分的情况,特别是走、上楼、下楼这3个动作数据非常相似,这给正确识别这3种人体动作带来了较大的难度。提出一种基于特征增强与决策融合的行为识别方法,通过对部分特征值进行增强处理和对多个分类结果进行决策融合来识别走、上楼、下楼这些难以区分的相似动作。实验验证,所提方法可克服由于加速度数据的相似性而导致的动作识别正确率低、识别误差大的情况,有效提高人体行为识别率,且可在实际应用中实时识别人体行为动作。

Activity recognition via tri-axial accelerometer has been a research focus in the fields of sensor data processing and pattern recognition.In some cases,it is difficult to distinguish the similar acceleration data,especially the data of walking,ascending stairs and descending stairs,which makes it difficult to recognize those three activities correctly.In this paper,a method of activity recognition based on feature enhancement and decision fusion was proposed for recognizing similar activities such as walking,ascending stairs and descending stairs,which is implemented by enhancing a part of features and decision fusing several classification results.Experimental results show that the method can overcome the situations of low correct recognition rate and high recognition error due to the similarity of acceleration data,and effectively improve the correct recognition rate of human activity and distinguish human activities in actual applications in real time.