EV-TTC is a high-speed Time-To-Collision (TTC) estimation system created using a fast multi-temporal scale event representation and a slim dilated convolution network. This repository provides the complete pipeline for TTC estimation, including C++ ROS2 nodes, neural network training, and dataset preparation.
vid.mp4
-
High-Speed Multi-Temporal Scale Filter
- Runs with a latency of 3.3ms at 75 Million Events Per Second on a Jetson Orin NX 16GB with events from 720x720 image resolution.
-
EV-Slim Neural Network
- A lightweight, high-speed network optimized with TensorRT for minimal latency.
-
$T^2CEF$ Dataset- A new dataset created from the M3ED dataset, consisting of ground truth optical flow and TTC for event camera sequences.
This folder contains the C++ ROS2 implementation for real-time TTC estimation. For more details, refer to the ev_ttc README.
This folder contains the PyTorch Lightning pipeline for training and evaluating the EV-Slim neural network. For more details, refer to the model README.
This folder contains scripts for preparing the $T^2CEF$ dataset from the M3ED dataset. For more details, refer to the TTCEF README.
@ARTICLE{ev_ttc,
author={Bisulco, Anthony and Kumar, Vijay and Daniilidis, Kostas},
journal={IEEE Robotics and Automation Letters},
title={EV-TTC: Event-Based Time to Collision under Low Light Conditions},
year={2025},
pages={1-8},
doi={10.1109/LRA.2025.3565150}}
If you use
@INPROCEEDINGS{m3ed,
author={Chaney, Kenneth and Cladera, Fernando and Wang, Ziyun and Bisulco, Anthony and Hsieh, M. Ani and Korpela, Christopher and Kumar, Vijay and Taylor, Camillo J. and Daniilidis, Kostas},
booktitle={IEEE Conf. Comput. Vis. Pattern Recog. Workshop},
title={{M3ED}: Multi-Robot, Multi-Sensor, Multi-Environment Event Dataset},
month={July},
year={2023}}
