Behavior recognition method and system based on skeletal joint point division and level division

Abstract

The invention belongs to the field of computer vision, particularly relates to a behavior recognition method, system and device based on skeletal joint point region division and hierarchical division,and aims to effectively improve the accuracy of behavior recognition and reduce the number of network layers. The method includes the following steps that each frame image of an input video is obtained, and skeletal joint points are extracted from each frame image; for each frame image, the extracted skeletal joint points are divided and classified to corresponding divided human-body regions, corresponding feature representations are obtained through graph convolution operation, and a first-layer feature representations set is obtained; for each frame image, according to the corresponding human body regions, based on the first-layer feature representations, the number of the joint points is reduced layer by layer through pooling and graph convolution until a feature vector is obtained through multi-layer aggregation, and the feature vector is input to two full-connected layers to get a behavior category. The accuracy of behavior recognition is improved, and the training speed and thedetection speed are increased.

Description

technical field [0001] The invention belongs to the field of computer vision, and in particular relates to a behavior recognition method and system based on skeletal joint points in different regions and in different levels. Background technique [0002] In the field of artificial intelligence, there is a skill called human behavior recognition, which is a basic technology for many applications such as intelligent monitoring, human-computer interaction, and robots. Take the smart nursing care for the elderly in nursing homes as an example. Through real-time detection and analysis of the behavior of the elderly, the intelligent system can judge whether the elderly are eating and taking medicine normally, whether they are maintaining the minimum amount of exercise, and whether there are abnormal actions (such as falling), and give timely reminders to ensure that the elderly The quality of life of patients will not be reduced, and the workload of caregivers can be reduced at th...

AI Technical Summary

Problems solved by technology

However, there are two problems in the current graph convolution method: one is that the entire graph convolution process is too flat, and only the local information of each joint point is learned, and the differences between the five regions of the human body cannot be learned. There are conclusions to prove that , many behaviors can be identified according to the information between the five regions of the human body, and learning the information between the five regions of the human body has an important impact on the classification results of behavior recognition; the second is that the most critical issue between graphs is information flow, This is a function that neither CNN nor RNN has

Images