基于单位统计曲率特征匹配的红外目标检测方法

A Method Based on Unit Statistical Curvature Feature Matching for Infrared Target Detection

摘要:

针对机器人在隧道施工环境中，由于阴暗强光等因素导致红外图像中的目标检测失败，提出一种基于单位统计曲率特征匹配的红外目标检测方法。采用最小二乘法对目标的曲面进行拟合，根据拟合曲面计算出目标中各像素的高斯曲率和平均曲率，使用曲率代替梯度构造图像特征描述符并建立曲率平面，根据曲率分布的密度将其划分为多个单位区域，对每个单位中的像素使用统计信息来生成稳定的单位统计曲率特征矩阵，通过计算矩阵之间的欧氏距离得到目标的相似性，识别红外图像中待检测的目标。对该算法与现有其它算法对标准图像数据集和实际施工隧道中的栈桥的检测准确率进行对比评价，结果表明，该算法的检测准确率最高，满足了工程上隧道机器人行进中识别栈桥的使用需求。

Abstract:

For the robot in the tunnel construction environment,due to factors such as dark and strong light,the target detection in the infrared image fails,a method of infrared target detection based on unit statistical curvature feature matching is proposed.The least squares method is used to fit the surface of the target,and the Gaussian curvature and Mean curvature of each pixel in the target are calculated according to the fitted surface.The curvature is used to replace the gradient to construct the image feature descriptor and establish the curvature plane.According to the density of the curvature distribution,the curvature plane is divided into multiple unit areas,and use statistical information for the pixels in each unit to generate a stable unit statistical curvature feature matrix,by calculating the Euclidean distance between the matrices to obtain the similarity of the target,it can find the target to be detected in the infrared image.This algorithm compares and evaluates the detection accuracy of standard image data sets and the actual construction tunnel trestle with other existing algorithms.The result shows that the detection accuracy of this algorithm is the highest,which meets the needs of tunnel robots in engineering to identify trestle bridges.