The Object Matching Technology For Multi-camera Under Training Scene Of Electriferous Jobs

Recently, with the appearance of public security problems and the development of video image processing technology, real-time video surveillance system has been made great use of in transportation, security and energy, etc. However, since these systems usually use single-camera way to achieve object detecting and tracking, it’s difficult to follow the object when it is covered. As a result, the camera collaborative network made of multi-camera can increase the viewing angle and bring us more rich video information, which can help solve the object cover positioning and other issues.This thesis aims at the vital function of multi-camera collaborative technology in real-time intelligent monitoring system, focuses on the key technology about object matching for multi-camera. It also works out the way object matches which based on axis extraction and information fusion of multi-view for moving object and has solved the object matching problem in certain extent caused by covering. The research content mainly contains the following points:(1) A method of axis extraction based on human body symmetry is raised. Namely, the axis of moving body is extracted because the human body is almost symmetrical about the axis. This principle is similar to the mean value theorems, so this method has excellent robustness while noise jamming is suppressed effectively. Moreover, this thesis aims at training platform of electricity work and works out another method for axis extraction which combines with safety helmet detection and the use of color futures and the accuracy is improved ultimately.(2) Based on the first point, another matching method of axis extraction for moving object which mixes multiple points of view information together is put forward. The key is to establish a geometric model according the mapping relationship between the public planes. The matchup relationship in different visual ranges of the same time can help achieve the matching of multiple objects. This method makes great use of the geometric information and the objects can be matched accurately.(3) Researches are achieved to make sure the accuracy of both methods above in the training platform of electricity work condition. The relevant results are given out in each situation of single-object, multi-object and covering object, which shows the method of axis extraction for moving object of multiple cameras has both excellent accuracy and robustness.