|
3D Pedestrian Trajectory Prediction using Deep Learning from Kinect Data
|
Akbar Jafari   , Ali Hosseininaveh * , Mojtaba Mahmoodian |
| K.N.Toosi University of technology; |
|
|
Abstract: (587 Views) |
Pedestrian movement trajectory prediction is one of the most important challenges in machine vision and it has attracted the attention of many researchers. In order to predict the movement path of moving objects such as pedestrians or cars, it is better to use the return networks that have the ability to retain the necessary information from the past path. Short and long-term memory networks (Long Short Term Memory-LSTM), which is a type of recurrent network, has the ability to retain information for a long time. Therefore, to predict the position of the pedestrian from the image data (including three green-red-blue bands), LSTM network is used after normalization. Since image data is based on two-dimensional space and pixels, therefore, the proposed LSTM networks for prediction are in two-dimensional space.
But considering that the real world is three-dimensional and one of the most important factors in pedestrian behavior is the distance between humans and other obstacles, especially moving obstacles. Therefore, in order to better predict the situation of the pedestrian, it is necessary that the proposed model be closer to the real world and the prediction of the situation of the pedestrian should be done in three-dimensional space. Therefore, in this research, from the RGB-D data obtained at the Polytechnic University of Lausanne, Switzerland. (EPFL) has been used to predict pedestrian position and 3D-LSTM network has been proposed to predict pedestrian position in metric three-dimensional space. The most important feature of this network is the prediction of the third dimension, which has a great impact on the decisions of the robots and their movement path. The obtained results show that the accuracy of predicting the three-dimensional position of the pedestrian is almost equal to the two-dimensional state and it also predicts the information of the third dimension.
|
|
| Keywords: Pedestrian Trajectory, Trajectory prediction, deep learning, LSTM Network |
|
|
|
Type of Study: Research |
Subject:
Aerial Photogrammetry
Received: 2023/06/10 | Accepted: 2024/05/26 | ePublished ahead of print: 2025/03/17
|
|
|
|
|
|
|
| Send email to the article author |
|
|
