The attention assigned to the driving task must be matched with its demand of safe driving, in order to explore the distraction intervention strategies of in-vehicle secondary tasks. An experimental vehicle was driven in naturalistic driving conditions to acquire real-time traffic data and videos of the road ahead. A prediction model was established to predict the driving task demand based on those real-time data. Participants assessed the driving task demand directly from short videos, verified the effectiveness of the prediction model, distraction intervention strategies under different driving task demand were proposed. The results showed that the consistency of driving task evaluation and prediction assessment is about 83%, there is no big difference, such as high forecasting demand and low evaluation requirements. Distraction intervention strategies based on real-time prediction of driving task demand can provide methods and technical support for the driver’s distraction management.