对红外序列图像中运动弱小点目标的关联检测问题进行了研究。在多假设检测的基础上采用“K/N”规则检测算法, 并引入带有可信度信息的判决融合, 其中, 可信度取自噪声概率分布函数。该方法把多假设检测与“K/N”规则检测算法有机地结合起来, 克服了多假设检测运算量大, 难以实时处理的缺点。引入可信度信息避免了当序列图像中某帧突然出现一个强的噪声脉冲时, 使得穿越该点的可能轨迹的统计量增大, 导致算法判断为出现多个虚假目标的情况。对信噪比较小的点目标, 特别是在有个别帧点目标被云层完全遮挡住的情况下, 该方法仍然有很高的检测概率; 最后, 从理论上详细分析了该方法的检测性能。理论分析和实测数据结果有效地验证了检测算法的可行性。

The association detection of dim small moving object in infrared image sequences was studied.Based on multi-hypothesis “K/N” regulation was adopted and credibility was introduced as noise probability distribution function.Multi-hypothesis and “K/N” regulation were organically combined which could overcome such problems of the former as large computation cost and inability for real-time processing.Credibility is helpful for improving the robustness of detection and the method has high detection probability even when there is noise pulse and cloud coverage.Theoretical analysis and real data test prove the feasibility and effectiveness of the method.