针对复杂干扰环境下多目标跟踪面临的目标数量时变、运动模态复杂及 “虚假目标” 干扰突出，导致跟踪精度显著退化的关键问题，本文提出一种置信度驱动的自适应高斯混合概率假设密度（Confidence-Adaptive Gaussian Mixture Probability Hypothesis Density, CA-GMPHD）滤波算法。该算法通过创新设计四类协同自适应机制，对传统 GMPHD 滤波的核心环节进行优化：似然层基于量测置信度动态调整测量协方差，提升高置信度量测与目标的匹配精度；先验层动态修正探测概率与杂波强度模型，降低低置信度量测被误判为真实目标的风险；融合层在权重更新过程中嵌入置信度幂权因子，强化置信度先验信息与几何一致性约束的融合效能；结构层基于全局平均置信度自适应优化剪枝与合并阈值，实现强杂波场景下滤波分量数膨胀的有效抑制。为验证算法性能，构建包含多传感器多目标及箔条云、角反射器两类典型强干扰源的仿真场景，量测置信度由检测器输出概率经映射函数生成。实验结果表明，相较于标准 GMPHD 算法，所提 CA-GMPHD 算法在全局均方根误差（RMSE）与最优子模式分配（OSPA）两项核心评价指标上均有明显降低，在保持计算效率的同时，显著提升了复杂干扰环境下多目标跟踪的精度与鲁棒性，具有重要的理论意义与工程应用价值。

To address the key problem in multi-target tracking under complex interference environments—where the time-varying number of targets, complex motion modes, and prominent "false target" interference lead to significant degradation of tracking accuracy—this paper proposes a Confidence-Driven Adaptive Gaussian Mixture Probability Hypothesis Density (CA-GMPHD) filtering algorithm, which optimizes the core links of traditional GMPHD filtering through the innovative design of four types of coordinated adaptive mechanisms: specifically, the likelihood layer dynamically adjusts the measurement covariance based on measurement confidence to improve the matching accuracy between high-confidence measurements and targets, the prior layer dynamically modifies the detection probability and clutter intensity model to reduce the risk of low-confidence measurements being misjudged as real targets, the fusion layer embeds a confidence power weight factor in the weight update process to enhance the fusion efficiency of prior confidence information and geometric consistency constraints, and the structure layer adaptively optimizes the pruning and merging thresholds based on the global average confidence to effectively suppress the expansion of the number of filter components in strong clutter scenarios. To verify the algorithm performance, simulation scenarios including multi-sensor, multi-target, and two types of typical strong interference sources (chaff clouds and corner reflectors) are constructed, where the measurement confidence is generated by mapping the detector output probability through a mapping function, and experimental results show that compared with the standard GMPHD algorithm, the proposed CA-GMPHD algorithm significantly reduces both the global Root Mean Square Error (RMSE) and Optimal Subpattern Assignment (OSPA)—two core evaluation metrics—while maintaining computational efficiency, remarkably improving the accuracy and robustness of multi-target tracking under complex interference environments and possessing important theoretical significance and engineering application value.