轮廓不变特征在待机飞行器识别中的应用

Application of Contour Invariant Features in Recognition of Standby Aircrafts

飞行器拍摄到的待机飞行器图像常出现旋转、尺度、仿射等畸变,同时噪声等影响会使目标轮廓部分缺失。针对这个问题,提出了一种轮廓不变特征,并将其应用于待机飞行器识别当中,以分割出来的物体灰度图像为基础,利用椭圆拟合方法进行方向归一化,提取全局轮廓特征;根据轮廓中的关键点位置将轮廓划分为上下左右4部分局部轮廓,提取局部轮廓特征,将其当作神经网络的输入参数,利用神经网络作为分类器,达到识别物体的目的。设计了两组目标识别对比实验。实验结果证明此方法在噪声污染、轮廓提取不完整的情况下,仍能得到较高的识别率,优于传统的矩特征等方法。

The ground standby aircraft images taken by the aircraft are often affected by the distortion of rotation, scaling, affine etc, and the noise effects can make partial of target contour missing. To solve this problem, a new feature, entitled contour invariant feature, was proposed and applied in recognition of standby aircraft. Based on the object gray image obtained through segmentation, the ellipse fitting method was utilized to normalize the direction of the targets, and the global contour features were extracted. Then, the contour was divided into four parts according to the key points coordinates, and the local features were extracted. The global and local features were served as an input vector of the trained neural networks to distinguish whether the source image was the destination image. Two groups of target recognition comparative experiments were made. The results show that this method can obtain high recognition rate in spite of the noise pollution or incomplete contour extraction, which has better performance than state-of-the-art algorithms such as moments feature.