Dynamic Time Warping (DTW) has been a very efficient tool in matching two time series and in past much work has already done in modifying DTW so as to enhance its efficiency and further broadening its application areas. In this paper we are proposing an enhanced version of DTW by calculating mean and standard deviation of the minimum warping path because of which the efficiency of DTW increased in detecting different human activities. We also introduce a new fusion of DTW with Histogram of Gradients (HOG) as it helped in extracting both temporal and spatio information of the activity and this fusion has worked very effectively to depict human activities. We used Random Forest as a classification tool giving highest accuracy of 88 % in weizMan dataset.
展开▼
