Improved Kalman filtering algorithm for indoor multi-target positioning with unknown measurement loss probability

Indoor multi-target localization systems hold tremendous potential for emerging applications such as digital warehousing, coal mine safety monitoring, and intelligent hospital navigation. However, indoor environments are highly susceptible to interference caused by signal occlusions and dynamic environmental variations, which severely degrade positioning accuracy. Most existing localization approaches fail to construct noise models that accurately reflect real-world conditions and lack effective mechanisms to cope with measurement information loss, resulting in poor robustness and limited practical applicability. To address these challenges, this paper proposes an improved Kalman filtering algorithm capable of handling unknown measurement loss probabilities. Specifically, the measurement noise is modeled using a Student's t distribution to enhance robustness against outliers; a Bernoulli random variable is introduced to refine the measurement model; and the prior distribution of the measurement loss probability is characterized by a Beta distribution. A variational Bayesian inference framework is then employed to jointly estimate both the state variables and the measurement loss probability with higher precision. In addition, the algorithm integrates absolute and relative measurement information to further improve localization accuracy and robustness. Comprehensive experimental results demonstrate that the proposed method outperforms five state-of-the-art algorithms, achieving a 21.02%–72.19% improvement in positioning accuracy, the average positioning accuracy of the proposed algorithm achieves 34.04 cm, thereby validating its superior performance and strong applicability in complex indoor environments.