抄録
The recognition of human actions is one of the most important areas in computer vision. It is a very difficult research area due to complexity of understanding human actions from video. In this paper, we study on the recognition of human actions based on modified Motion History Image (modified-MHI). The MHI is an extracted image from the video of human actions, which contains the necessary motion information. We proposed a way to better represent MHI, which is immune to noises due to camera movement and motion of unwanted objects. Edge detection of moving objects and finding dominant motion portion are important factors for efficient human action recognition. Finding dominant motion portion is a very challenging task because of the change of background intensity at daylight and movement of camera. We have obtained better MHIs from edge detection. The obtained MHI templates are processed afterwards to find the binary patterns and motion gra- dient features. These features are exploited for action recognition by using Support Vector Machine (SVM). We have obtained 98.1% accuracy in classifying actions on Pedestrian Action Dataset which proves the robustness of our proposed method. This dataset is comprised of 8 different actions and each action is performed by 20 different subjects.