A Study On Real-Time Visual Metrology

Body behavior recognition is a very important task of criminal investigation, Safety monitoring and behavior analysis. The height of a human is a important parameter for recognition. For example, a unexpected change of height means tumble. in addition, as the image resolution is not high enough or the camera is far away from body, the face recognition becomes more difficult, then the height of body will become more important to the body information. Compared with other measurement methods, such as laser and radar. The measurement technique based on image/video just takes on the scene for image and video information simply, and it can be measured without contacting, repeatedly at any time. Those advantages make this technique be useful to study. Compare to the static images, videos have two advantages:the first is that the result is more accurate and reliable because we can calculate the result using probability statistics; the second is that we can get the result real timely so that we can response to the emergencies in the scene. In a word, the research of dynamic human body height of the video has important theoretical significance and application value.Regarding the dynamic human body height measurement of video, firstly, we calculate related parameters in Camera scene mainly including two ways:calibrated and uncalibrated camera. Secondly, we, extract human body region in videos, which can be called foreground extraction. There are a lot of foreground extraction methods, mainly including two ways, adjacent frames difference and background modeling. Finally, we extract feature points on the human body’s area by using the calculated parameters, so we can build height model, and solve the height of the body.This paper established the height measuring system. The first is to calculate parameters of the fixed camera in the scene, and we test the measuring methods of calibration and uncalibrated camera. Then in the foreground extraction, we realized the method of the adjacent frames’difference, background difference and the gaussian mixture model. On the effect and speed of the foreground extraction, we study hard. The measure method is concluded, which is the method of the gaussian mixture model. Then we optimize the method of the gaussian mixture model. In the case of guaranteeing the test results, the test’s speed increased from223milliseconds per frame to40milliseconds per frame, which achieve the real-time demand of the system measurement. Finally In the aspect of feature point selection, we refer to the existing feature points extraction algorithm, and put forward a new kind of foot feature points extraction algorithm. It combines the spatial characteristics of human body’s posture, avoids the real-time tracking of your feet, can be able to accurately locate the foot feature points, and measures the height fast by using the height of the human body model.