基于改进Fast MBD显著性检测和多特征融合匹配的靶纸区域快速检测算法

为了解决基于机器视觉的自动报靶系统快速、准确定位靶纸区域的问题，通过对靶纸图像的颜色和形状特性分析，提出一种基于改进Fast Minimum Barrier Distance显著性和多特征匹配的靶纸区域快速检测算法。该算法在原始Fast Minimum Barrier Distance显著区域提取算法的基础上，引入局部区域对比度先验和形状先验作为显著性区域提取的补充准则。同时，为了判断提取到的区域是否包含靶纸，再引入多特征匹配算法。首先，分别对图像边界连通先验、局部区域对比度先验和形状先验进行量化，形成距离图、对比度图和形状图，再结合三者分割出显著性目标区域，然后提取分割出的各目标区域的多种特征，并将其进行特征融合，最后利用1-范式将得到的目标特征与模板特征进行匹配，把匹配结果小于阈值的目标视为靶纸。在400张包含靶纸图像数据集上的实验结果显示了该算法的有效性。同时，在华为海思平台上，该算法处理速度能达到30帧/秒，足以证明该算法的实时性。

A Fast Detection Algorithm of Target Sheets Based on Improved Fast MBD Saliency Detection and Multi-Feature Fusion Matching

In order to solve the problem of automatic target-scoring systems, which is finding target sheet regions quickly and accurately from image, we propose a fast detection algorithm of target sheets based on Improved Fast Minimum Barrier Distance Saliency Detection and Multi-Feature Fusion Matching by analyzing the color character and the shape character of target sheets. We introduce local regional contrast prior knowledge and shape prior knowledge that are the compensation extraction criteria of salient regions to the origin Fast MBD Saliency Detection. Meanwhile, to determine whether the extracted region contains a target sheet, we introduce Multi-feature Fusion Matching. Firstly, we quantify image boundary connectivity prior knowledge, local regional contrast prior knowledge and shape prior knowledge respectively to calculate distance map, contrast map and shape map. Then we incorporate the three maps to segment salient regions. When we get salient regions, we extract their multi-feature to measure similarity with saved model feature by L1 norm. Finally, we regard the salient region whose similarity measuring value is less than threshold as target sheet. Experimental results on the dataset that has 400 images containing target sheets show that the algorithm we proposed is effective. Meanwhile, it is enough to prove real-time that the speed of the algorithm we proposed will achieve 30FPS at Hua Wei HiSilicon platform.