Human body behavior identification method and device

Abstract

The embodiment of the invention provides a human body behavior identification method and device. The method comprises the steps that a graphical user interface (GUI) is provided, the GUI is used for displaying a control panel and a display panel, the control panel comprises a network loading button, an image loading button and an identification button, and the display panel is used for displayinga to-be-identified image and a behavior identification result; if it is monitored that the network button is triggered to be loaded, a pre-trained GoogLeNet model is loaded; if it is monitored that the image loading button is triggered after the GoogLeNet model is loaded, displaying the image in a display panel after the image to be identified is received; and if the image is displayed on the display panel and the trigger identification button is monitored, the image is inputted to be identified into the GoogLeNet model, a behavior identification result is outputted and the behavior identification result is displayed in the display panel. When the GUI is constructed, the human body behavior recognition process can be displayed directly and the purpose of repeated recognition can be achieved only by loading the model once during application.

Description

technical field [0001] Embodiments of the present invention relate to the technical field of image recognition, and more specifically, to a method and device for human behavior recognition. Background technique [0002] In recent years, human action recognition has gradually become a very active research topic. Action recognition remains a challenging task due to the complexity of actions, such as different body wear and habits leading to different observations of the same action, camera motion in external environments, lighting changes, shadows, and viewpoints. s project. Most of the research in behavior analysis is based on video, while there are relatively few studies on static images. However, in recent years, research on human behavior recognition in static images has gradually gained attention. [0003] There is a method of "poselet activation vector" in the prior art, see the reference: "Action recognition from a distributed representation of pose and appearance", ...

AI Technical Summary

Problems solved by technology

This method is simple and does not need to consider human body detection, object detection, etc., but the recognition accuracy is low because it cannot distinguish the subtle gap between different behaviors and the interaction between people, scenes, and objects.

Images