This paper proposes to investigate the use of active devices, able to anticipate for hazardous situations, by using low cost sensors. In this approach, the risk of rollover is considered thanks to the Lateral Load Transfer (LLT) metric, able to characterize gradually the dynamic mass repartition on the vehicle, without using expensive cell forces. In order to account for variable conditions, an observer algorithm is used in order to adapt on-line the grip conditions thanks to a dynamical yaw model. Once adapted, this model supply the dynamic variable influencing the evolution of LLT. This metric can then be estimated thank to a second partial dynamic model considering the roll plane. Thanks to the vehicle parameters (mass, elevation of centre of gravity), the risk of rollover may be accurately computed.