Study On The Normalized Cross-correlation Wavefront Gradient Processor Based On CPU

With the development of the technology, adaptive optical(AO) system has been widely used in the fields of large telescopes, laser wavefront correction, and human eye optometry and so on. But with the promotion of application, the correction object and the using environment of AO system become more complex. This presents a serious challenge to the wavefront controller. As the front end of wavefront controller, the performance of gradient processing directly affects the correction effect of the whole AO system. Therefore, this paper did research on gradient algorithm optimization by analyzing characteristics of different gradient algorithms and different hardware platforms.First of all, the calculation of the centroid algorithm, the absolute difference algorithm and the normalized cross-correlation algorithm was analyzed in this paper, and several simulation experiments were designed to compare the performance of these three algorithms in different detection conditions. Although the normalized cross-correlation algorithm has large computation, it has strong noise resistance, and can resist the interference from the pseudo spot to the target detection. Besides it can also be used for extended target detection and flickered target detection. According to the characteristics of the normalized cross-correlation algorithm, the features of some commonly used hardware platforms were analyzed, and a multi-core CPU was chosen as the realization platform of the normalized cross-correlation gradient algorithm.Secondly, this paper analyzed the tasks of gradient calculation with the characteristics of normalized cross-correlation algorithm, and carried out optimized research on gradient algorithm. On the basis of dependence and independence between data in gradient calculation, thread-level parallelism was implemented by pipelining and parallel processing, so that the processing time was shortened. Then, according to the characteristics of the multi-core CPU, parts of code in the gradient processing application were optimized by AVX instructions, and data-level parallelism was implemented, which further enhanced the efficiency of the normalized cross-correlation gradient calculation.Then, the time of the normalized cross-correlation slope detection was tested on the Intel(R) Core(TM) i7-3770- k quad-core computer and Windows 7 operating system, while the template was 9*9 pixel and the Hartmann image had 400 subapertures with 15*15 pixel. The test results showed that the gradient detection time of optimized procedure was about 340 μs, which was about 29% of the time spent by the non-optimized one.Finally, the optimized normalized cross-correlation slope algorithm was used in an AO system, and a closed loop experiment was performed. The experimental results showed that the optimized normalized cross-correlation gradient algorithm could detect wavefront gradient effectively, and had good ability to resist the interference of pseudo flare.This paper realized the normalized cross-correlation gradient algorithm on a general multi-core CPU. This implementation can make the AO system meet the needs of a variety of work scenarios. It has such features as nice commonality, high transportability, and strong expand-ability, so that it can increase processor cores to meet the need of a larger AO system.The study in this paper provides effective technical foundation for extending the application of wavefront gradient processing in AO system. Beyond that, the study has important research value and practical engineering significance.