基于Hi3559A平台的地物分割目标检测方法并行设计

Parallel design of object detection method for ground object segmentation based on Hi3559A platform

为了在复杂背景下检测出红外低慢小目标,提出一种基于地物分割的目标检测方法。使用边缘检测算法得到地物以及目标边缘信息,利用图像列信息得到每一列的临界点,并求出一个异形区域;使用引导滤波算法,通过求得的异形区域来剔除假目标。实验结果表明,该方法在复杂背景下对小目标的检测有较强的的稳定性和鲁棒性。将基于地物分割的红外低慢小目标检测方法移植到Hi3559A开发平台上,使用并行设计,将多线程和基于ARM架构处理器扩展结构NEON技术来优化检测方法。为了验证方法的实效性,将未优化和优化后的方法进行比较,结果表明,优化后的方法在速度方面比未优化方法快近2倍,仿真测试结果表明,在640×512分辨率大小的图像上能够满足100 Hz的检测要求。

In order to detect infrared low altitude small target for images with complex backgrounds,a new method for guided filter based on ground object segmentation is proposed.The edge detection algorithm is applied to get the edge information of the target and background features,and use the image column information to get the critical point of each column,and find a special⁃shaped area to eliminate fake targets after guide filter.Experimental results show that the method has strong stability and robustness to small target under complex backgrounds.The infrared low⁃slow and small target detection method based on ground object segmentation is transplanted to the Hi3559A development platform,using parallel design,multithread and NEON technology based on the ARM architecture processor extension structure are used to optimize the detection method.In order to verify the effectiveness of the method,compares not optimized and optimized methods.The optimized method is nearly 2 times faster than the unoptimized method.The simulation test results show that the 100 Hz detection requirement can be met on the 640×512 resolution image.