Visual Perception for Autonomous Driving
in Day and Night

Our Goal
Toward Next Generation of Vision Technology

Our Team
Our team pursuits the Psychological Safety, because Psychological Safety calls for open communication & empathy.  

New project page

Research Area
  • Visual Recognition
    • Only Lidar based Object Detection
    • RGB-Lidar/RGB-Thermal based Object Detection
    • RGB or Thermal Image Enhancing for Visual Recognition
    • RGB/Thermal based Scene Understanding
  • Multi-Sensor Integration
    • Sensor Calibration
    • Sensor Synchronization


A New Perspective on Thermal Imaging Enhancement using Convolution Neural Network
IEEE International Conference on Robotics and Automation (ICRA) 2018, submitted
[project page] Patent Pending (KR 10-2016-0100058). Winner in KINPEX 2016.
KAIST Multi-spectral Recognition Dataset in day and night
Transactions on Intelligent Transportation Systems (T-ITS), Accept after minor change.
[project page] 

Multispectral Transfer Network: Unsupervised Depth Estimation for All-day Vision
Association for the Advancement of Artificial Intelligence (AAAI) 2018. (To appear)
Received Gold Prize 23th HumanTech Paper Award, Samsung.
Patent Pending (KR 10-2017-0023475)

Evaluation for Mobileye Technology
Technical Report
Thermal Image Enhancement using Convolution Neural Network
In Proc. of IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2016.
[project page] 1st place in NVidia Deep Learning Contest 2016. 

Two-Phase 3D Convolution Neural Network for Object Detection
Technical Report
[project page] 
Fast Multiple Objects Detection and Tracking Fusing Color Camera and 3D LIDAR for Intelligent Vehicles
In Proc. of IEEE International Conference on Ubiquitous Robots and Ambient Intelligence (URAI), Oct 2016.
[project page] 
  Thermal-Infrared based Drivable Region Detection
 In Proc. of Intelligent Vehicles Symposium (IV), June 2016.
 [project page] 
Geometrical Calibration of Multispectral Calibration
In Proc. of  International Conference on Ubiquitous Robots and Ambient Intelligence (URAI), 2015.
[project page] Patent Pending (KR). 
Low-cost Synchronization for Multispectral Cameras
In Proc. of IEEE International Conference on Ubiquitous Robots and Ambient Intelligence (URAI), Oct 2016.
Multispectral Pedestrian Detection: Benchmark Dataset and Baseline
In Proc. of Computer Vision and Pattern Recognition (CVPR) 2015.
[project pageFirst multispectral pedestiran dataset. 
All-Day Visual Place Recognition: Benchmark Dataset and Baseline
In Proc. of Computer Vision and Pattern Recognition Workshop on Visual Place Recognition in Changing Environments (CVPRW-VPRICE) 2015.
[project page] First multispectral place-recognition dataset

Going Work

Light and efficient Neural Network
Weakly supervised learning for pedestrian detection
SLAM with tunnel vision


Talk or Presentation

2016/09/xx - Luxir 
2016/08/10 - Hankook Tire
2016/06/09 - Volvo Korea
2016/04/04 - NASA
2016/03/22 - in CPE Lyon
2016/03/04  - Prof. Masayuki Tanaka 
2016/02/01 -  i3system
2016/01/13  - in Samsung Electronics
2015/07/27  - Dr. Kyunghwan An in ETRI  

[2015] This project was supported by the Ministry of Trade, Industry and Energy of Korea (MOTIE) and LIG-Nex1
[2016] This project was supported by the National Research Foundation of Korea(NRF) grand funded by Korea government(MSIP).

MOTIE: "Autonomous Emergency Braking System for Pedestrian Protection project"
LIG-Nex1: "Detection and Tracking Multiple Objects"

Special Thanks to the Dr. Kyunghwan An in ETRI & RCV members