A real-time warning model of vehicle-person collision risk based on minimum safe distance is a technology to protect the safety of pedestrians in traffic in the network connection environment. Based on the minimum safe distance theory, this paper proposed a real-time warning model of human-vehicle collision risk with dynamic coefficients. Firstly, a real-time warning model of human-vehicle collision risk was constructed. Secondly, a accident model of pedestrian with headphones was established. In this stage, typical models were selected to analyze the collision characteristics, based on which the appropriate human-vehicle distance algorithm was summarize and thus the margin of safe distance in different scenarios was determined. Finally, the warning effect of the model was calculated in the term of braking distances under different speeds, sections and road adhesion coefficients, and compared with the braking distances without the intervention of the warning system. The results showed that the model not only has high accuracy, but can effectively warn drivers and pedestrians so as to reduce the accident rate.