Research On Method Of Automatically Locating Ground Object And System Research

The task of locating a target in a video sequence has proved to be challenging. It has been an important branch of computer vision, which is significant both in theoretic area and in reality. Especially in military, it has become one of the core techniques of accuracy missile guiding. The object is not always the same as the template. On the one hand, there are often some scale, rotation, and grey value changes between the real object and the template. On the another hand, the object is not stable and it is always changing with the object movement such as rotation, scaling and translation.The traditional correlation matching method has poor performance with large biases in the presence of the object's large scale change and rotation angles. So ,in order to get more high precisely locating result, proposed a robust object locating method based on log-polar transform and affine transform which can locate the object misaligned by scale and rotation changes to the template in this paper. We combined the Log-polar transform and the affine transform to accurately locate the object images misaligned due to rotation and scale change. The Log-polar transform can furnish a good intial affine parameters estimation, even in the presence of arbitray rotation angles and a wide range of scale changes. And then the affine transform can locate the affine parameters accurately. And an object locating method based on the Log-polar transform and PCA to multiple object templates is proposed. The identical object will display different form with the changes of the imaging distance ,the imaging view and the object's pose. The normal image segmentation and character extraction method can't perform well in complex battlefield environment. And the multi-templates method is difficult to be used because of the large storage and high omputational complexity. We adopted PCA to compress the compress the large datas, and utilized the good character of the Log-polar transform can convert the scale and rotation change to translation change, and advanced a novel algorithm to recognize the image object automatically.