<p>The conventional phase plane method estimates the stability of vehicle systems using fixed inputs. This approach exhibits a clear inclination towards conservatism, potentially leading to unnecessary interference with the driver or failures in collision avoidance due to stringent limitations during emergency scenarios. To address the aforementioned issues, this study introduces a regression region constraint by extending the phase plane method. Subsequently, an emergency collision avoidance system (ECAS) is developed based on this regression region. According to the emergency degree of the scenario, ECAS system is divided into 2 modes: the immediate takeover mode and the risk monitoring mode. In the mode of immediate takeover, the system promptly substitutes the driver upon encountering obstacles and employs the collision avoidance algorithm based on optimal control and model predictive control to avoid collisions. When the vehicle states exceed the constraint of the regression region, the regression stability algorithm is triggered to make the vehicle regress to stability. In the risk monitoring mode, the driver avoids collision by himself, and the system monitors the vehicle states. The regression stability algorithm replaces the driver once the regression region constraint is reached. The performance of two intervention modes was validated based on the dSPACE HIL platform and G29 driving simulator correspondingly. The results suggest that the immediate takeover mode can be qualified for collision avoidance conditions with the longitudinal distance of 33.0 m at the speed of 30 m/s, while the shortest distance the existing studies can deal with is 37.5 m, and the risk monitoring mode can also successfully avoid unnecessary interference when the driver is competent.</p>