Behavior recognition method based on light stream information

Abstract

The invention discloses a behavior recognition method based on light stream information, and belongs to the technical field of computer image processing. The behavior recognition method comprises the following steps that: single-behavior video sequence light stream diagrams are obtained; the moving direction mode of each light stream vector of the light stream diagrams is determined through the direction information of all light stream vectors in a neighborhood; in addition, amplitude values are determined by utilizing median filtering, then, different moving amplitude modes are divided according to the amplitude values, and the moving mode of each light stream vector is determined through the moving direction mode and the moving amplitude move of the light stream vectors; each light stream diagram is divided into a plurality of rectangular blocks according to rows and lines; the moving modes of all of the light stream vectors in each rectangular block are counted for forming moving mode histograms; the histograms of all of the rectangular blocks of each light stream diagram are subjected to cascade connection for forming feature vectors of the light stream diagrams; the feature vectors of all of the light stream diagrams are subjected to cascade connection again to obtain current behavior vectors; and different obtained behavior feature vectors are trained and recognized on the basis of SVM. The behavior recognition method has the advantage that during the behavior recognition, the inhabitation capability on complicated backgrounds is high.

Description

technical field

[0001] The invention belongs to the technical field of computer image processing, and mainly relates to behavior recognition in video images. Background technique

[0002] Human action (action behavior) recognition is an attractive and challenging problem in computer vision. Human behavior recognition refers to the analysis, understanding and recognition of the movement patterns and postures of the human body. It is an advanced visual research on the analysis and understanding of human movement in videos, and it belongs to the category of artificial intelligence. Visual analysis and understanding of human motion and action recognition can be applied in many fields, such as: motion capture, video surveillance, human-computer interaction, environmental control and monitoring, sports and entertainment, etc. Especially in terms of video surveillance, with the decreasing cost of security monitoring equipment such as cameras, video surveillance systems can be wide...