Prediction on trajectories of pedestrians becomes more essential than ever in the era of the development of autonomous vehicles.
This project compares performances of different methods on predicting x-y coordinates based on sequential time steps.
Inputs: x-y coordinates of sequential steps of a person
Output: x-y coordinate of the next step of this person
Methods: LSTM, GRU, KNN with linear regression
The programs are implemented in Tensorflow, Python 2.7
The idea comes from Social-LSTM experiment of Stanford University.
The paper named Social LSTM: Human Trajectory Prediction in Crowded Spaces
This project is not the duplicate of the experiment in the paper, nor a reimplement of the algorithm mentioned in the paper. The original code of the lab can be referred in this website: https://github.com/vvanirudh/social-lstm-tf