Development of Pedestrian Route Prediction Model Using Deep Learning Based on 2D LiDAR Sensor Data

초록

This study uses deep learning to model various features and patterns of walking behavior beyond the limitations of traditional theoretical systems that interpret the complex walking patterns of pedestrians with rules and mathematics. The first pedestrian movement path prediction model transforms pedestrian movement path data into a time series cell method, constructs the walking path process of the pedestrian to occupy the cell, and learns and predicts the deep neural net (DNN) model. In order to predict the bypass path of the pedestrian due to the second resistance, the pedestrian bypass path is extracted from the circulation path that has been freely detoured in the pedestrian’s path and decreasing the resistance value by increasing the number of times by +1. The walking path of the pedestrians are constructed based on the cell and integrated to learn and predict the DNN model. The model performance is tested by comparing the two models presented in this study to the Social Force Model. K-means clustering was performed to compare the path diversity of two existing prediction models to DNN. As a result of the comparison, it is confirmed that a reasonable level of path prediction is possible because the number of representative paths is the same.