Video-based non-linear on-line expression pre-detection method and device

Abstract

The invention discloses a video-based non-linear on-line expression pre-detection method and device thereof. The method includes performing preprocessing on video data and extracting feature information of each frame of image; video segments with different lengths are extracted from the training samples, and the nonlinear mapping between the segments and the detection function is constructed. Therepresentation of fragments is based on multi-instance learning, fully mining the effective expression information, and constructing a constraint pair for every two video fragments according to the size of the mapping value to obtain training data; at each time point, the OKMEFD model is updated with all the constraints of the current sample. By minimizing the experience loss and model complexityon the current training data set, the model is optimized with stochastic gradient descent algorithm to obtain the online updating criteria of the model, and the nonlinear expression pre-detection function at each time point is further obtained. The pre-detection method is that the data is read frame by frame to the test sample and the corresponding detection value is output. When the value is greater than the set threshold value, the expression event is considered to be pre-detected. The invention realizes the pre-detection before the end of the expression video. The invention effectively improve the timeliness of expression recognition.

Description

technical field [0001] The invention relates to an expression pre-detection method and device, belonging to the interdisciplinary technical fields of facial expression recognition, video processing, and event pre-detection. Background technique [0002] After entering the 21st century, with the rapid development of artificial intelligence technology, computer technology and related disciplines, the demand for human-computer interaction has become increasingly strong. Faced with the deepening of related research and huge application prospects, facial expression recognition based on video In the past ten years, it has become a research hotspot in the field of artificial intelligence and machine learning, and has made great progress. Although related algorithms emerge in endlessly, many theories are still incomplete, and there are almost no mature commercial results. There is a process for the generation and disappearance of expressions. The traditional recognition method is t...

AI Technical Summary

Problems solved by technology

[0006] The technical problem to be solved by the present invention is the complexity of the facial expression pre-detection problem. The present invention proposes a video-based nonlinear online facial expression pre-detection method OKMEFD (online kernel multi-instance learning for early facial expression detection), which further improves the pre-facial expression detection. Validity of detection

Images