PERFORMANCE IMPROVEMENT OF PEDESTRIAN TRAJECTORY PREDICTION BY PREDICTION ERROR CORRECTION UTILIZING LSTM

Abstract

In this paper, a pedestrian trajectory prediction method which introduces a prediction error correction is proposed. Specifically, an improvement of both accuracy and precision of prediction is targeted by utilizing two prediction models in parallel, each using a pedestrian trajectory as its input. These two models respectively predict a tentative coordinate from the pedestrian trajectory and predict the prediction error including in the coordinate prediction. Finally, the predictions of the two prediction models are integrated to calculate the prediction results of the proposed method. The two prediction models are created by utilizing Long Short-Term Memory (LSTM). As a result, the prediction accuracy and precision are expected to be improved, because the correction can be made based on the pedestrian trajectory. An experiment was conducted using pedestrian trajectory data which was obtained from drone videos. The experimental results show that the proposed method improves the accuracy and precision of the pedestrian trajectory prediction.