The invention discloses a mobile terminal based pre-warning system for distractive walks, comprising a model training module and a real time detection module. The model training module conducts pre-treatment and model training based on the large amount of training data it has collected to generate a sample model for real time detection. The real time detection module, through the use of the mobile terminal, acquires the sample and conducts the pre-treatment. Based on the sample model generated by the model training module, the real time detection module conducts separate detections to the acquired samples. According to the results of the separate detections, decisions and judgments are made so that warnings can be made to a distractive user who walks out of a sidewalk. According to the invention, it is possible to make warnings to a user in advance before the user walks towards a dangerous zone. Without extra hardware arrangement, effective warnings can be made to the mobile terminal of a distractive walker, and the previous detection results can be effectively utilized. With high accuracy, the pre-warning system consumes less energy and can make judgments in advance.