This period of time is the image processing of the extraction of the ball. Maybe it seems simple, but you know, the extraction of the ball here is to use in both eyes, so you must have a certain precision.
It is still very simple now, but it is already possible to set the general position of the ball, but a small part of the pixels of the ball sometimes can't split. The reason is that I am using only when it is H value, and the s value is auxiliary threshold segmentation technology. In fact, it is like a paper mentioned in a paper, that is, the L value comparison In high cases, the H value is divided, and it can achieve a very good effect. When the motion is blur (this, the color is diffused), or the light is too dark (for a small ball, the lower part of the small ball will compare the dark), the H value cannot be obtained, At this time, use the S value. According to my judgment, this theory is right. The reason why the corresponding code has not been written yet, how should the L value set, and how to divide different parts of the same picture in the same picture, use different L values, then how, these have not thought understand.
Below is the steps I do, it is very simple, but it will definitely there will be a problem.
Mainly used by regional growth methods. First, a fast HLS transform is performed, and a fast transformation is performed with reference to a paper. Then, since the method is used, the seed point is selected, and the threshold segmentation is performed according to H and S, search in the entire picture, to get the most likely point, when doing seed points, the seed point at this time Still only initial choice, the next step is to remove the small messy seed point (with morphological corrosion calculations), through this method, you can remove the relatively small interference point. Then, the seed point of this to use this will then perform regional growth. After the operation, the expansion and corrosion of the region will be performed so that each area will compare complete, there is less empty. When this is finished, there will be more than one area. At this time, the method based on region-based counting is used, and the purpose is not counting, but the ball is found (also using a priori knowledge on green venues). After this is processed, the effect is ok, but it is also improved.
