基于边沿方向特征的地面时敏目标识别方法

A Novel Ground Time-Sensitive Target Recognition Method Based on Edge Orientation Features

地面时敏目标机动性强,目标姿态无法准确预知导致识别出现困难。传统SIFT特征对旋转、尺度、光照等畸变有很好的抑制作用,但是其只能在较小的视角变换范围内起作用,当视角变换较大时SIFT识别率较低。针对此问题,提出了一种基于边沿方向特征的地面时敏目标识别方法。首先利用积分图像及Haar小波模板计算图像梯度场,以模值极大值点作为特征点;然后将满足距离约束的两点组合成为特征点对,利用互相校验的方法,增强了特征点对的独特性;最后将梯度方向作为匹配依据,实现目标识别。实验结果表明,给出的识别方法具有可行性和有效性,可以在大角度视角变化中保持稳定,在45°范围内均能实现目标的正确识别,具有较强的鲁棒性,优于SIFT算法。

The mobility of the ground time-sensitive targets is very strong,and the target attitude information can not be accurately predicted,which may lead to identification difficulties.SIFT algorithm exhibits great performance under a variety of image transformations,such as rotation,scale,illumination and other distortions,but it can only work within a small range of perspective transformation.When encountering large perspective transformation,the recognition rate of SIFT becomes lower.To solve this problem,a ground time-sensitive target recognition method based on edge orientation features was proposed.Firstly,integral images and Haar wavelet template were utilized to calculate the image gradient field,taking the points with maximum module value as feature points.Two points that meet the distance constraints were combined to form a pair of feature points,enhancing the peculiarity of the feature point pairs by using mutual calibration method.The gradient orientation was taken as the matching basis to achieve target recognition.Experimental results show that the recognition method is feasible and effective,which can keep stable within a wide range of perspective transformation.In the experiment,targets within the perspective of 45＆#176;can be identified correctly.The method has strong robustness to the perspective transformation,and performs better than SIFT algorithm.