Abstract:

In order to solve machine vision reduce the effects of background and noise on the video feature extraction, and to improve recognition results. We propose
 a human action analysis method by combining the local space-time interest points and global accumulated edge image feature based on the study of human action representation and the full consideration of the advantages and disadvantage of the local and global features. First, local space-time points of interest and global accumulated edge image feature are extracted from the video sequences. Then we use the weighted dictionary to combine them together organically, finally we use the minimums distance method for human action analysis and identification. This method can effectively obtain the spatial characteristics of human, human edge contour and the human movement trends and intensity, the experimental show that our method is faster and gains a higher recognition accuracy generally inceased by 2% to 5%.

Key words: space-time point of interest, accumulated edge image, action analysis