|
Human activity recognition is an important topic in computer vision , one of the researchers in recent years, much attention is also a research direction , can be widely used in intelligent video surveillance, intelligent robot , motion analysis and other areas. However, due to non- rigid motion of human motion , and there is the appearance , physique , exercise habits and other differences , making identification difficult human behavior , so far, no one can identify common behavioral model , most of the studies in a particular scene is carried out . The purpose of this project is to realize human walking, running , jumping and bending over four acts of recognition , and in the process of identifying some of the algorithms used in research. Since moving target detection is the first step in human behavior recognition , test results have a direct impact on behavior recognition result , this paper first human movement detection algorithm is studied . Were used inter-frame difference method , statistical averaging method and Gaussian Mixture Model For three methods for moving object detection experiments , and the advantages and disadvantages of each method are analyzed , and finally selected this method as a Gaussian mixture model for moving target article detection algorithm . For the detection results of the image , the paper has conducted a noise removal , and according to characteristics of the human body detected by the moving target for a determination whether the human body . For the detection of human movement , the paper has conducted a feature extraction and feature extraction algorithm is studied . Two methods were used to extract features : one is based on Hu moments feature extraction algorithm, the other is based on Fourier descriptors feature extraction algorithm. Then, based on these two feature extraction algorithm , the establishment of two human model : Based on the Hu moments mannequin and feature improvements based on Fourier descriptors mannequin . The final step is the identification of human behavior , the paper first four acts to be identified define keyframes , and then uses a sampling intervals key frame extraction algorithm for key frame extraction, and then the extracted key frames for feature extraction established two keyframe-based human model libraries, respectively, based on the last two complete human body model library behavior recognition, and recognition results were compared and analyzed .
|