Space-time significance region detection-based human body behavior analysis method

Abstract

The invention discloses a space-time significance region detection-based human body behavior analysis method. According to the invention, a data set is adopted to rain a Faster R-CNN model. A multi-channel video is input and a single-channel video is segmented into video image frames. The segmented video image frames are subjected to target detection by adopting the Faster R-CNN model. A target detection result is analyzed and a target candidate box is re-calculated. The single-channel video is subjected to box matching for constructing a motion vector field. Based on the motion vector field,the motion vector of the region of interest is calculated. A foreground remarkable motion region is selected based on a probability calculated by utilizing a Gaussian mixture model. According to the target candidate box and the remarkable motion region, a space-time significance region is synthesized. The target space-time significance region is subjected to feature sampling and feature pretreatment. The target space-time significance region of the video is subjected to encoding and pooling. The space-time significance region of the video is subjected to human body behavior analysis and identification. An obtained analysis and identification result is written into a space-time significance region box. According to the invention, the categories of human body behavior activities in a video can be reasonably analyzed.

Description

technical field [0001] The invention relates to the technical field of video analysis, in particular to a human behavior analysis method based on spatiotemporal salient region detection. Background technique [0002] Human behavior analysis based on video belongs to the field of video analysis. Because human behavior has variability such as posture deformation, perspective change, time difference, and scale, and video images are easily affected by camera shake and illumination changes, video-based human behavior Analytics becomes a hard problem to solve. [0003] The methods of human behavior analysis can be roughly divided into three categories: methods based on human body model tracking, methods based on optical flow histograms, and methods based on local spatiotemporal features. The methods based on human body model tracking require the extraction of accurate human body templates, and the robustness of this method is relatively poor. Optical flow histogram-based methods...

AI Technical Summary

Problems solved by technology

The advantage of static image target detection is that it can use the saliency of spatial structure to extract various targets. The disadvantage is the lack of time series analysis, resulting in inaccurate expression of significant moving areas of interest, and lack of correlation between specific targets (such as human bodies or objects) Semantic analysis, which affects the accuracy of human behavior recognition

Images