As sensor types diversify and random finite set (RFS) algorithms advance, the capabilities of tracking platforms have significantly improved, resulting in increasingly complex tracking scenarios. Initially, tracking objectives focused on employing a limited set of sensors to achieve multi-target tracking. But the focus has shifted to more specific objectives of surveillance, such as reducing the uncertainty of target estimates to an extent that permits accurate weapons deployment. In this case, traditional multi-sensor management systems, which balance tracking performance with system resources, are inadequate for these specific control objectives. Therefore, this paper adopts a covariance control approach, integrating with the δ-generalized labeled multi-Bernoulli (δ-GLMB) filtering algorithm from RFS, to select sensor combinations on the basis of discrepancies between the expected covariance matrix and the predicted covariance of each target. Using the recursive filtering, we present a multi-sensor selection algorithm based on recursive δ-GLMB filtering, named MSSR-δ-GLMB algorithm. Simulation results show that our MSSR-δ-GLMB algorithm saves nearly one third of resources while maintaining tracking accuracy.