Design Of Processing Algorithm For Hartmann Spot Image Based On FPGA

Shack-Hartmann Wavefront Sensor has been widely used in adaptive optics system. The calculation of the light spot centroid is the first processing operation of the image data to rebuild the wavefront, and it is also a very important operation. Due to the increasing CCD resolution, the image data size to be processed has greatly increased. Therefore, it is very difficult for the traditional data processing methods to meet the real-time requirements of the system, such as methods based on PC and DSP. As a kind of programmable hardware platform, FPGA is capable of parallel computing, which makes it an ideal choice to process images. The connected component labeling algorithm and barycenter algorithm are adopted to calculate light spot centroid. The main feature of the algorithms is that the amount of data and computation is very large, but there is no complex arithmetic computation and the algorithm structure is relatively simple. So it is suitable to implement on FPGA platform.Firstly, the parallel processing method is adopted in this paper. One frame image is divided into two sub-images and they can be processed synchronously in two processing cores within the FPGA. This processing method is based on the pixel unit and it takes one clock period to process one pixel for every processing core. However, in a spot image, the proportion of light spot pixels is very small, and most of them are invalid background pixels. For example, there are only about 100 spots in a 1k×1k image and each spot only contains 20 pixels or so. Thus, most of the calculation time is wasted on the scanning of the background pixels. Then a fast algorithm is proposed by improving the scanning mode during the labeling processing on the basis of the spot image characteristics. This processing method is based on the cell unit with the size of 4x4 pixels. Thus a cell can be seen as a big pixel and 16 pixels can be processed within one clock period. However, this method will have a problem when the distance between two light spots is less than 8 pixels. So a complete processing flow of problem judgment and re-labeling is designed, and it is also one of the key contents of the fast algorithm. The simulation result indicates that the processing speed of the fast algorithm is much higher than the parallel processing algorithm.