Abstract
Action recognition is a challenging task in computer vision. In this paper, a new skeleton-based action recognition, which uses joint trajectory images and vision feature to model human action, is proposed. First, Openpose is applied to extract human skeleton data from RGB Camera. Human action in a single video is represented as the trajectory of human(s) skeleton joints in an image space. Inspired by the work of optical character recognition, the problem of action recognition can be regarded as the problem of trajectory image recognition.HOG combined with SVM method is applied to recognized action in this paper. Finally, in the experiment, the recognition performances are analyzed and compared. Our work can be helpful for researchers to try to use the joint trajectory images for action recognition.
Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
