A method, device and system for virtual reality interaction are applicable to the technical field of virtual reality. The method comprises the steps that image information from an automatic capture camera is received at first, and sensing information which is collected by at least one collector and sent by a corresponding virtual scene client side is received; according to the image information and the sensing information, spatial position information and posture information of all the users are acquired; the spatial position information and the posture information are transmitted to all the virtual scene client sides, so each client side can render a virtual scene and display the virtual scene to a local user according to the spatial position information, the posture information and perspective information of the local user; the local user is one of all the users; and a group of optimal data and a group of inertia data are integrated, so the users can be positioned when the users aresheltered mutually.