Conventional pedestrian trajectory prediction cannot predict the short-term social conflict in advance especially in crowd areas.Thus

an algorithm is proposed to predict the short-term social conflicts between pedestrians

where the pedestrian trajectories in image can be obtained via the YOLOv3 network from the original video

and then send to Kalman filter after format conversion to estimate the future trajectories via linear regression.A judgment model is designed for the pedestrian state detection and potential conflict judgement

which finally give the social conflict prediction.The experimental results show the proposed method can accurately predict the upcoming conflicts of pedestrians

and give accurate judgments on various states.Besides

it can process the image frames 42 and 5 times faster than other existing algorithms S-LSTM and S-GAN in the case of dense pedestrians

which verify its advantage in computation cost.