In mission-critical wireless sensor networks surveillance applications, a high detection rate coupled with a low false alarm rate is essential. Additionally, fusion methods can be employed with the hope that aggregation of uncertain information from multiple sensors enhances the quality of surveillance provided by the network. This paper investigates the following fundamental problem: what is the best way to deploy a finite number of unreliable sensors characterized by uncertain readings in order to satisfy the user detection requirements. Unlike prior efforts that rely on simple fusion schemes, we use the Dempster-Shafer theory to define a generic evidence fusion scheme that captures several characteristics of real-world applications. The fusion-based uncertainty-aware sensor networks deployment problem is formulated as a binary non-linear and non-convex optimization problem that is NP-hard, and an efficient heuristic using genetic algorithms is investigated. The effectiveness and efficiency of the proposed approach are evaluated using both simulations and experiments. The obtained results demonstrate the appropriateness of the evidence fusion model that considers in a meaningful way the information on the quality of sensors decisions as well as the reliability of these sensors along with their uncertain and imprecise decisions. Also, the proposed approach outperforms state-of-the-art deployment strategies.