Rollover accidents are a serious and too frequent incident at many locations on the road. Especially in the condition of trucks or Sport Utility Vehicles (SUV) travelling at high speeds require a greatly reduced speed when meeting the exit ramps and tight curves. The usual cause of rollovers is driving behavior, typically excessive speed while cornering which adversely affects the stability of the vehicle. Sudden or severe changes in direction can create a potential risk to rollover. By the time drivers see or feel something wrong, it is usually too late to prevent a rollover. Although some papers discuss many methods to eliminate the rollover phenomenon, it could not provide early warning for the driver. These systems usually work in the condition of slipping or near rollover. To overcome the problem, an alternate approach is to incorporate an image-based detection technique with rollover prediction model. This paper estimates the maximum rollover threshold speed in real time by using the vehicle speed, acceleration and roll angle. The goal of this paper is to provide more than 90% precision of lane radius detection. In our test results, the average error of image recognition is within ± 6 m and affects 3% error of the maximum rollover threshold speed. It detects the lane radius effectively. The predicted value of the maximum rollover threshold speed is verified by measured slip angle. Compared to traditional methods, we could offer 2 or 3 seconds early warning before the vehicle rollover occurs, and hence the proposed approach is potentially suitable for application in rollover prevention systems.