A Human Action Recognition Method Based on Lie Group Features and Convolutional Neural Networks

Abstract

The invention relates to a human action recognition method based on Lie group feature and convolutional neural network, and belongs to the field of computer pattern recognition. The method includes: S1: data acquisition, using the Microsoft somatosensory device Kinect to extract human skeleton information, and acquiring the motion information of the experimenter; S2: extracting Lie group features, adopting a rigid limb transformation to simulate the relative three-dimensional geometry of each limb of the human body The Lie group skeleton representation method of relationship models human actions as a series of curves on the Lie group, and then combines the correspondence between Lie groups and Lie algebras, and uses logarithmic mapping to map the curves based on the Lie group space to the Lie algebra-based space. S3: Feature classification, integrate Lie group features and convolutional neural network, use Lie group features to train convolutional neural network, and let convolutional neural network learn and classify Lie group features, so as to realize human action recognition. The present invention can achieve good identification effect.

Description

technical field [0001] The invention belongs to the field of computer pattern recognition, and relates to a human action recognition method based on Lie group features and a convolutional neural network. Background technique [0002] With the rapid development of science and technology, more natural human-computer interaction has become an increasingly urgent need for people. People are more eager for computers to think and understand external input signals like the human brain, and to understand human daily behaviors. Easily and naturally communicate with the computer. [0003] Human motion recognition refers to a practical technology that takes digital images or video signal streams as objects, and obtains human motion information through image processing and automatic recognition. Due to the variability of human movements, camera movements, changes in light intensity, differences in human body types, and differences in human environmental conditions, the research on huma...

AI Technical Summary

Problems solved by technology

[0008] In view of this, the object of the present invention is to provide a human body action recognition method based on Lie group features and convolutional neural network, which greatly overcomes the interference of traditional technologies on external environment changes and human body shape changes, etc., and can be compared It overcomes the defects that some action recognition methods based on traditional Euclidean space cannot simulate and express the spatial complexity and geometric relationship of human actions; at the same time, this method can better deal with the similarity between actions and the high variability between classes; In terms of computing cost and recognition effect, using convolutional neural network to process features can not only learn and classify features well, but also reduce computing cost to a large extent; the recognition accuracy is high

Images