A method for recognizing human behavior in a video

Abstract

The invention relates to the field of artificial intelligence, more specifically, to a method for recognizing human behavior in a video. The main algorithm idea of the invention is as follows: firstly, the input video segment is pretreated, a six-parameter affine model is used to simulate the motion of a camera, and the motion trajectory is obtained through the affine model; Secondly, the video key frames are extracted based on temporal saliency and spatial saliency. Then the improved dense trajectory is extracted from the key frames, and the two-stream convolution neural network is selected as the feature extractor to extract the depth learning feature from the key frames. The extracted features are normalized, and the features are fused based on the video representation, which fuses thevideo representation based on the depth feature and the video representation based on the improved dense trajectory feature. The invention integrates the improved dense trajectory feature IDT and thedepth learning feature designed manually, and can more effectively mine the complementary information of the two features and the behavior pattern in the video, so as to obtain better effect in videohuman body behavior identification.

Description

technical field [0001] The present invention relates to the field of artificial intelligence, and more specifically, relates to a method for human behavior recognition in videos. Background technique [0002] The current research interest in human behavior recognition has shifted from simple behavior recognition in a well-controlled shooting environment to more realistic behavior recognition in unconstrained environments such as story movies, sports broadcast videos, and home videos. The research difficulty in this case The main reason is that the behavior of the human body is easily affected by many factors such as angle changes, camera shake, illumination, background occlusion, and rapid and irregular movement of background clutter. Therefore, it is particularly important to extract effective features of human body movements in such videos. At the same time, different motions may have similar performances in videos, which also makes it a main research content to construct ...

Images