This paper is concerned with the problem of two targets tracking over sensor networks. A heterogeneous sensor network framework in considered, in which two types of sensors are employed (denoted as x-type and y-type sensors, respectively). Sensors with x-type can only measure information from target x, while sensors with y-type can only measure information from target y. Sensors update their tracking estimations by using only local information collected from neighboring ones. The network topology is random and governed by a continuous-time Markov process, whose states represent possible modes of interaction topologies of sensors. To solve the two targets tracking problem, a novel distributed tracking protocol in the form of consensus-based estimators is delicately developed. Then, tracking performance analysis against the effects of random network topologies is carried out as well as a criterion for designing desired tracking protocol parameters is derived such that the two targets tracking can be achieved. A numerical example is given to demonstrate the effectiveness of the proposed theoretical results.