Real-time expression feature extraction and identification method

Abstract

The invention relates to a real-time expression feature extraction and identification method. The method includes the following steps that a kinect is used for tracking a human face and extracting face action unit information AUs and feature point coordinates FPPs from real-rime video data; then the two types of the feature information are subjected to parallel processing, feature data in feature channels of the AUs and the FPPs are identified in advance through a 7element 1-vs-1 classifier group, obtained pre-identifying results are stored in a cache for emotion confidence statistics, results with the highest confidence coefficient are expression identification results in the channels, and finally, the results of the two feature channels of the AUs and the FPPs are integrated to obtain a final expression identification result. According to the real-time expression feature extraction and identification method, the problem of non-ideal speed and accuracy of common expression identification methods can be solved, and real-time high-accuracy expression identification can be achieved on the basis of face features that are extracted by the kinect.

Description

technical field [0001] The invention relates to a real-time expression feature extraction and recognition method, in particular to a method for recognizing expression based on facial motion units extracted by Kinect and feature point coordinate information. Background technique [0002] In the past two decades, with the rapid development of artificial intelligence and pattern recognition technology, the research on facial expression recognition technology has received extensive attention from researchers. Today, its applications cover a wide range of fields, such as gaming, security, affective computing, and human-computer interaction. [0003] Traditional expression recognition methods still have great limitations. In terms of recognition accuracy, most of the research is based on ordinary cameras, which can only collect flat 2D images. However, the human face is an object that exists in three-dimensional space. Using 2D images to describe 3D human faces will inevitably c...

AI Technical Summary

Problems solved by technology

[0005] Traditional expression recognition methods need to be further improved in recognition accuracy and speed

Images