基于极线约束的图像目标识别可信度评价方法

Reliability Assessment of Object Recognition Based on Epipolar Constraints

在图像辅助的惯性末制导系统中,提出了一种基于极线约束的图像目标识别可信度评价方法.当惯导系统给出的摄像机姿态或图像中目标的像素点位置存在误差时,为了平衡极线约束方程需要引入一个冗余量(residual),该冗余量可由惯导系统给出的摄像机两位置的相对运动及相应图像中目标点的像素位置计算得到.当惯导系统的姿态误差小于0.3°时,冗余量可以用来评价图像目标识别的可信度.理论推导及仿真实验证明,冗余量作为图像目标识别可信度的评价指标,具有良好的特性.

Based on epipolar constraints, a method of reliability assessment of object recognition is proposed in vision-aided terminal guidance system which used inertial navigation system （INS）. When sensor noise comes into play, the extra value had to be added to maintain equality of the epipolar constraint is called a residual, which can be computed from the information of INS and images. When the error of the camera attitude~ which proposed by INS, is lower than 0. 3 degree, the value of residual can reflect the error of object in images. The results of theoretical deduction and experiment indicate that the residual, as reliability assessment of object recognition, has several desirable properties.