Moving object recognition algorithm with high-definition video data suffers from large computation complexities and slow speed. With NVIDIA Tesla K20,c GPU,a method of accelerating the template matching target tracking algorithm with the heterogeneous system integrated with CPU and GPU was proposed. The parallel algorithm was designed by three optimizing means:constant memory,the internal memory of SMX and the brief calculation of correlation coefficient. Finally,the program was coded on compute unified device architecture and tested. The results show that the designed algo-rithm can obviously improve the real-time performance of the algorithm and guarantee the recognition effect.%针对大数据量导致模板匹配目标识别算法计算时间长,难以满足快速检测的实际需求问题,在采用最新NVIDIA Tesla GPU构建的CPU+GPU异构平台上,设计了一种模板匹配目标识别并行算法。通过对模板图像数据常量化、输入图像数据极致流多处理器片上化和简化定位参数计算3方面优化了并行算法,并对算法进行性能测试。实验表明,该算法在保证识别效果的同时实时性明显提高。
展开▼