|
In recent years, both at home and abroad in the field of video image analysis of a very important branch—movement object recognition, behavior analysis technology has achieved rapid development, its application has become increasingly broad areas, including intelligent video surveillance systems, virtual reality system and digital video encoding technology. Human movement analysis is not to need people to intervene, or only rarely required intervention, through analyzing the camera recording the video sequence to realize the people detection, locating and tracking, and on this basis to obtain the body posture and movement parameters, and further understanding of semantic analysis and behavior.In this paper, theoretical and practical application of point of view, focusing on the movement body’s recognition, focusing on the current most popular for locating the head of torso mask matching algorithm, as well as edge detection and smoothing algorithm related to recognizing torso, on the movement human recognition methods to explore and research, and on basis of method carried out by the system experiments application and verify. The main research content and innovation are as follows:1. On the current use of more of the head locating algorithm①Color parameters based on the human face②parameters of head shape information, based on the comparison and observation, found that the Micilotta’s head locating algorithm has the advantages for instance simple, intuitive and easy to realize,using the algorithm the head and torso to form a Chinese character Tu shape characteristics to locate the head.2. On the human body with more pronounced characteristics of corrosion-frame image is to be detected edge, using a variety of detection methods integrated treatment. Edge detection algorithm used in the main have:Canny Detect operator, Roberts Detect operator, Sobel Detect operator, Prewitt Detect operator, Kristh Detect operator, Guass-Laplacian Detect operator, one of Canny algorithm the outline of a clear, less burr edge, the effect is better. 3.After the edge detection, high frequency signals is generated. By gauss low-pass filter suppressing these high frequency signals, thus eliminated noise, smoothed image edges.4, On the basis of the target detection and tracking experiment system,some function is optimized, such as:the elimination of pseudo-objective, added a number of useful features, such as:movement target screenshot,complements the character template.5. According to the above researches,this paper was designed based on the movement human recognition experimental system, algorithm in this article are verified and tested by the experiments,the result show that the use of methods of recognition and extraction of the above-mentioned effect is prominence, research work can achieve the desired goal.
|