The safe transition between autonomous and manual control is essential to human-machine cooperative driving systems. This paper proposes a safe and efficient control transition framework based on the definition and estimation of the autonomous system’s capability of operating the vehicle. To describe whether the autonomous controller can safely operate the vehicle, the vehicle state subsets (safe, recoverable and unsafe regions) are formulated using sets of both scenario-specific and general safety-related variables. By introducing these state sub-sets and safety envelopes, the proposed framework can determine whether a controller steers the states across the safety bounds using reachability prediction. A control transition decision logic is designed to achieve a safe switch between autonomous controller and manual driving by online monitoring of possible violations of safety properties. The proposed framework is evaluated using a variety of simulated complex scenes. Experimental results show that the proposed framework can effectively improve the safety of human-machine cooperative driving systems.