Human Action Recognition Method Based on Structured Feature Map

Abstract

The invention discloses a human behavior recognition method based on a structuralized characteristic pattern. The method comprises the steps that a space-time interest point is extracted from a sample, and a characteristic vector is generated; the characteristic vector is set as an initial value, parameter training of a hidden condition random field model is completed by utilizing a preset function, and an element with intermediate semantic information and a spatial connecting relation of the element are learned; the sample is divided into timing sequence units, characteristic patterns are established inside the timing sequence units, and the characteristic patterns are mapped in a characteristic space; the timing sequence units are connected through a timing sequence relation, a chain-shaped graph is established, and a timing sequence graph kernel is brought up; the timing sequence graph kernel is utilized, and to-be-recognized behavior video is classified and recognized. By adopting an HCRF to complete the learning on the element and the incidence relation of the element, by using the characteristic patterns to complete the description of a video sequence, and by using the timing sequence graph kernel to complete the matching of the characteristic patterns in space and time, behavior sequences of different lengths and velocities can be matched, and the method can be used for the behavior of an individual and also be used for the interactive behaviors of many people.

Description

technical field [0001] The invention belongs to the technical field of visual information processing, and in particular relates to a human behavior recognition method based on a structured feature map. Background technique [0002] Now is an era of information explosion. Information of various contents, forms, and carriers comes one after another, filling our lives, and more than 80% of the information people get is visual information. Faced with such a huge amount of information, it is increasingly unable to meet the needs only by manually processing information. Therefore, it is urgent to research and develop computer-related capabilities that can replace manpower. Human behavior analysis based on vision is a very active research field. At the same time, human behavior recognition has extensive application value in human-computer interaction, intelligent monitoring, sports performance analysis, content retrieval and other fields, and has become a popular research field of ...

AI Technical Summary

Problems solved by technology

Although the method of local features + word bag model is simple and effective and does not need to track the human body, this method ignores the spatial-temporal correlation of local features

In addition, at present, the research on human behavior recognition technology mainly focuses on simple single-person behavior recognition, while the analysis of more practical multi-person interactive behavior is rarely involved.

Images