In response to data anomaly issues encountered during the detection and tracking of objects in complex space environments, conventional tracking filters based on single-step prediction often suffer from track interruption or erroneous tracking. To enhance the continuous and accurate prediction and tracking of moving targets, Moving Horizon Estimation (MHE) is introduced as an online track quality auditor. By comparing the tracker’s output trajectory with a dynamic model within a sliding time window, and given that MHE is fundamentally an optimization problem seeking an optimal solution, the cost of finding this solution—namely, the Optimization Residual—increases when there is a mismatch between the observed trajectory and the dynamic model. The established “verify-and-repair” framework can achieve precise detection and localization of anomalous data when this residual signal peaks. The performance of the MHE-based anomaly detection module is validated through simulations. Experimental results demonstrate that the module achieves high precision and recall in detecting anomalies in typical scenarios of target ID switches and drifts, providing robustness for object tracking and outperforming traditional methods based on innovation checking. This work presents a technical approach for building next-generation intelligent tracking systems capable of self-diagnosis and error correction.