A Lightweight Framework for Fast Trajectory Simplification

Introduction

This repository holds source code for the paper "A Lightweight Framework for Fast Trajectory Simplification".

Environment Preparation

• Python 3.8.8 (Recommend Anaconda)
• CentOS 7.0
• Pytorch == 1.8.1+cu111
• A Nvidia GPU with cuda 11.7
• Please refer to the requirement.txt to install all required packages of Python.

Datasets Description & Preprocessing

Prepare your own trajectory data with the following format and each trajectory is split by "\n". Put the data file into ./datasets and run python preprocess/createDataset.py. The preprocessed datasets will be split into training sets, validation sets, and testing sets which is stored in ./datasets folder.

[[116.51172, 39.92123], [116.51135, 39.93883], [116.69171, 39.85182]]

[[116.69171, 39.85184], [116.6917, 39.85184], [116.6916, 39.85177]]

python preprocess/createDataset.py

Besides, prepare the roadmap data named as "edge.edgelist" which is corresponding to the trajectory. Run the code like python preprocess/node2vec_main.py --input datasets/edge.edgelist --output datasets/beijing.emd to generate your own road node embedding vector with semantic information.

python preprocess/node2vec_main.py --input datasets/edge.edgelist --output datasets/beijing.emd

Note: the roadmap data of tdrive datasets can be downloaded in repo.

Running Procedures

Hyperparameters

You can create a config file like model_configs/s3.yamland specify your own hyperparameters for better performance, including learning rate, loss function, min/max compression ratio and some network layer dimension.

Training

Run s3.py, the generated models will be stored in the folder ./checkpoints automatically, and you can pick one model with best performance as your model to do some experiments.

python models/s3.py --config ./model_configs/camera/s3.yaml

Note: you can choose any config yaml file with the format like --config ./model_configs/camera/s3.yaml.

Evaluation

Run debug_outputs.py to generate the compressed trajectory and constructed trajectory and the output will be saved in the folder ./evaluation.

python generate/debug_outputs.py

Citation

If you use our code for research work, please cite our paper as below:

@article{,
  title={A Lightweight Framework for Fast Trajectory Simplification},
  author={Ziquan Fang, Changhao He, Lu Chen, Danlei Hu, Qichen Sun, Linsen Li, Yunjun Gao},
  year={2023},
}