Human Behavior Recognition Based on Bag-of-Words Model in Logarithmic Euclidean Space

Abstract

The invention discloses a human body behavior recognition based on a logarithmic Euclidean bag-of-words model, and belongs to the technical field of digital image processing. The present invention firstly divides the input video into fixed-length and overlapping video segments, and then divides each video segment into fixed-size and partially overlapping space-time cubes, and extracts gradient and optical flow feature covariance or Shape feature covariance, and use symmetric positive definite matrix dimensionality reduction method to reduce dimensionality of covariance matrix. The covariance matrix is ​​logarithmically transformed, and the upper triangular feature of the logarithmic covariance is extracted and transformed into a logarithmic Euclidean space vector. In the logarithmic Euclidean model, the bag-of-words model is used to model the behavior, and the spectral clustering is used to cluster the behavior features to generate a codebook, and the locally constrained linear coding technology LLC is used to encode the behavior features. Use nonlinear support vector machine to train and identify and classify behavioral features. The invention is used for human body behavior recognition, and its robustness is excellent.

Description

technical field [0001] The invention belongs to the technical field of digital image processing, and relates to relevant theoretical knowledge such as computer vision and pattern recognition, especially human body behavior recognition based on a logarithmic Euclidean bag-of-words model. Background technique [0002] Human behavior recognition is a research hotspot and difficulty in the field of computer vision. Its core is to use computer vision technology to automatically detect, track, and recognize people from video sequences and understand and describe their behavior. Human motion analysis and behavior recognition algorithms are the core content of human behavior understanding, mainly including video human detection, tracking moving human body, obtaining relevant parameters of human behavior, and finally achieving the purpose of understanding human behavior. [0003] Human behavior recognition methods are mainly used in intelligent monitoring systems to actively and real...

AI Technical Summary

Problems solved by technology

[0008] Human behavior recognition is affected by factors such as inter-class changes and intra-class changes of human behavior, behavior execution environment, camera position, and changes in human behavior in time and space, which greatly limits the improvement of recognition accuracy.

Behavior representation often cannot effectively integrate different features, reduce the strength of feature representation, and reduce external interference

Images