| In order to meet with the need of visual surveillance, this paper put forward a thought and means of object localization of rotatable single-camera in visual surveillance, based on machine vision in which a number of digit image disposal technology and pattern identification theory are used.With the relationship between the project position of the head of mobile object and the distance between object and camera, the man height model of fixed single-camera can be built which is proved in this paper. By this model, the two-dimensional position of object on the ground can be located effectively. Because the head can't be covered easily, this method can not only be used outside but can also be used inside with complex background. Experiments with simulated data show that the technique is simple and accurate. Then, for the change of the model because of the rotation of the camera, this paper proposed an update algorithm to obtain the model coefficient and the sample parameter which have been changed. The original parameters and the angel were combined in this algorithm to solve the update with two approach of pitching and pan respectively. The localization of object can be implemented by the model of new position which can be built by these parameters. In order to calibrate the camera accurately and efficiently, an accurate method of camera self-calibration based on the circle point and the vanishing point on the orthogonal directions is proposed in this paper. This method requires two or three image of a model plane with octagon and diagonals, because of the invariability of IAC in rigid motion and the relation between the absolute conic and orthogonally, the internal parameters of the camera can be calculated linearly. This method needn't any matching and physical measurement, and the process is simple. The result of experiments shows that this method is efficient and has good precision. The angelwas obtained beforehand by the method of quaternion. |
PDF Full Text Request