Abstract: To solve the problem of cooperative target observation for a multiple UAV (unmanned aerial vehicle) system with a complex communication condition, we propose a distributed state estimation method based on multi-agent consensus theory. Considering the loose communication structure and the complexity of the communication condition, we design an iteration mechanism with double-time windows, including a local prediction/update window and a consensus fusion window, to improve the filter precision of the distributed system. Then, we propose a filter method for distributed unscented information based on the consensus algorithm. The influence of the convergence property of the consensus algorithm on the estimation precision is analyzed theoretically, revealing the fundamental reason why the covariance error of the proposed method is greater than that of the centralized method. The Monte Carlo simulation experiments indicate that the proposed method is robust to complex network constraints, and has outstanding performance on average estimation error, average consistency error and average trace of the covariance matrix. This method can meet the requirement for the real time estimation of non-linear targets for a multiple UAV system under a complex network.