MOTION RECOGNITION USING SILHOUETTE IMAGES

Abstract

Human activity recognition has been a significant subject in computer vision because of its various uses or application, for example, video reconnaissance; human machine association and video recovery. One of the main troubles behind these applications is mechanically recognizing low-level activities and abnormal state exercises of concern. This Thesis shows a strategy of naturally distinguishing the activity finished by the human. The speeded up robust features (SURF) algorithm is made utilization of, for extraction of the key-focuses in both spatial and worldly area which is the augmentation of the SIFT finder/detector. The Spatio-Temporal Difference-of-Gaussian (STDoG) pyramid is basically fabricated which is additionally used to discover the maxima and the minima focuses which give the intrigue focuses. The key focuses are start in the x-y, x-t and y-t planes where x-y matches to the spatial plane, x-t and y-t planes take after to the transient spaces. Experimental result is lead on a video covering numerous activity recognition on KTH dataset.