Human Action Recognition Method using YOLO and OpenPose

Human action recognition is a technology that recognizes human behavior through images. This can be used in video surveillance to detect abnormal situations without human intervention. To increase accuracy of behavior recognition, it is considerable method to use consecutive images. However, it increases time complexity due to the large number of videos to be processed in real system. Thus, the lightweight model which can recognize human behavior with just one image is necessary. To this end, we propose a lightweight behavior classification model based-on neural network which utilize human joint information.