Abstract: The traditional trajectory prediction algorithm based on the actual position cannot dig the pedestrian intention information deeply, which leads to some defects of pedestrian trajectory prediction and active collision avoidance algorithm. In the proposed algorithm, the pedestrian image and location information were acquired based on vehicle-mounted sensors, and the motion features of pedestrian were identified based on convolutional neural network. The Kalman filter algorithm was used to obtain the predicted value of state estimation, and the predicted trajectory conforming to the pedestrian subjective intention was output. The safe distance models of different pedestrian paths were established based on the intersection characteristics of people and vehicles, and the active collision avoidance algorithm for pedestrian was designed based on the road pedestrian trajectory prediction. The experimental results show that the trajectory prediction algorithm based on motion feature analysis can predict the trajectory changes of pedestrian in advance and effectively guarantee the safety of pedestrians. The proposed trajectory characteristic classification can describe the intersection of pedestrian and vehicles under mixed environment. The active collision avoidance algorithm not only improves the safety of pedestrian and autonomous vehicles, but also ensures the smoothness of braking deceleration process and traffic flow.