Human activity recognition based on silhouette analysis using Local Binary Patterns

In this paper, we present a novel method for activity analysis in outdoor environment. Difficulties, such as low resolution, shadows, long distances, and segmentation problems, usually exist in outdoor environment. To deal with these difficulties, the activity width sequence image which maintains 3D features using 2D representation is exploited to represent the silhouette structure information of each frame and dynamic properties of activity. The activity width vectors are converted to the gray value successively according to the order in activity sequence and the gray image is formed in spatio-temporal space. We regard the activity width sequence image as the texture and choose Local Binary Patterns which is a powerful texture extraction operator to analyze the spatio-temporal pattern. The method is simple and effective. We test our method based on the outdoor database and the results are encouraging.