Thermal-infrared based Drivable Region Detection.
we propose a novel algorithm of drivable region  detection that suits thermal-infrared based images.

# Question (Jae Shin Yoon)
Personal Home page : Please click here.

The Paper is uploaded! (download)  [bibtex]


AbstractDrivable region detection is challenging since various types of road, occlusion or poor illumination condition have to be considered in a outdoor environment, particularly at night. In the past decade, Many efforts have been made to solve these problems, however, most of the already existing methods are designed for visible light cameras, which are inherently inefficient under low light conditions. In this paper, we present a drivable region detection algorithm designed for thermal-infrared cameras in order to overcome the aforementioned problems. The novelty of the proposed method lies in the utilization of on-line road initialization with a highly scene-adaptive sampling mask. Furthermore, our prior road information extraction is tailored to enforce temporal consistency among a series of images. In this paper, we also propose a large number of experiments in various scenarios (on-road, off-road and cluttered road).      A total of about 6000 manually annotated images are made available in our website for the research community. Using this dataset, we compared our method against multiple state-of-the art approaches including convolutional neural network (CNN) based methods to emphasize the robustness of our approach under challenging situations.

# The algorithm demonstration (Introduction + poster PPT) is uploaded!!! (download)

# The databases (5941 frames) of thermal-infrared based road scene are uploaded!!!

[ Database specification ]
:: You can download all the databases including ground-truth in below. 
:: Each frame is composed with three channels. First channel is ground-truth. Second or third channels are original thermal image.

    1.  Campus in Kaist (download)  - On road( 3547 frames).    Corresponding RGB images (download) (181 frames).

    2.  City (download) - Complicated road (1307 frames)

    3. Mountain (download) - Off road (1087 frames)

# The Video is uploaded!!! (download)