By considering the barrier-following motion and collision avoidance, a novel search method based on a simplified virtual-force model is proposed for multi-target search of swarm robots (SRSMT-SVF) in unknown environments with non-convex, convex and dynamic obstacles. The multi-target search behaviour of swarm robots in complicated environments is firstly decomposed, and a simplified virtual-force model is then formulated. Based on the proposed model, the motion control strategies of the individual robots under coordinated search and roaming state are designed to achieve real-time collision avoidance in searching process. Simulation results on different scale of swarm robot systems demonstrate that the proposed method can keep the individual robot with a good performance of collision avoidance, and can effectively reduce the collision conflicts between the robots and the environment as well as collision conflicts among the individual robots during the searching process. Moreover, compared with the extended particle swarm optimization (EPSO), the search time and energy consumption are reduced by at least 13.78% and 11.96%. The numeric simulation results demonstrate its effectiveness.