Seong Ik Cho

Methods to Detect Road Features for Video-Based In-Vehicle Navigation Systems

Understanding road features such as position and color of lane markings in a live video captured from a moving vehicle is essential in building video-based car navigation systems. In this article, the authors present a framework to detect road features in 2 difficult situations: (a) ambiguous road surface conditions (i.e., damaged roads and occluded lane markings caused by the presence of other vehicles on the road) and (b) poor illumination conditions (e.g., backlight, during sunset). Furthermore, to understand the lane number that a driver is driving on, the authors present a Bayesian network (BN) model, which is necessary to support more sophisticated navigation services for drivers such as recommending lane change at an appropriate time before turning left or right at the next intersection. In the proposed BN approach, evidence from (1) a computer vision engine (e.g., lane-color detection) and (2) a navigation database (e.g., the total number of lanes) was fused to more accurately decide the lane number. Extensive simulation results indicated that the proposed methods are both robust and effective in detecting road features for a video-based car navigation system.

Building Recognition for Augmented Reality Based Navigation System

In this paper, we propose a building recognition algorithm for the next generation navigation system which is to use captured video from the camera equipped in moving vehicle. After masking out the road area and vehicles on the road, we used edge segments of small size blocks within an image frame to remove background objects, trees, etc. Search range determination and edge tracing technique were also developed to determine the building area as the final step. The experiments show promising results such as 88.9 % detection rate on the average and 95.9% for navigation video without trees and signs. The proposed algorithm can be also applied to point-of-interest (POI) detection for various telematics applications.

Accurate dense stereo matching of slanted surfaces using 2D integral images

This paper presents an advanced algorithm providing accurate stereo correspondences of two frames through concise disparity gradient estimation at the per-pixel level and 2D integral images. The key contributions of this novel algorithm are twofold: First, combining an upright cross-based support region with disparity gradient estimation realizes the implicit construction of a 3D support region for each anchor pixel. This approach yields the disparity accuracy for slanted surfaces as well as fronto-parallel surfaces. Second, the 2D integral image technique leads to a speedup of matching cost aggregation in the implicit 3D support regions. The experimental results show that the proposed algorithm can successfully convey the correspondences of actual sequence of outdoor stereo images and Middlebury stereo images with high accuracy in near real time.

Traffic Light Detection Using Rotated Principal Component Analysis for Video-Based Car Navigation System

This letter presents a novel approach for traffic light detection in a video frame captured by an in-vehicle camera. The algorithm consists of rotated principal component analysis (RPCA), modified amplitude thresholding with respect to the histograms of the PC planes and final filtering with a neural network. The proposed algorithm achieves an average detection rate of 96% and is very robust to variations in the image quality.

Building detection in augmented reality based navigation system

In this paper, we propose a building detection algorithm for video based navigation system which overlays driving directions on the captured video from in-vehicle camera. We used edge segments of small size blocks within an image frame to remove background objects, trees, etc. Search range determination and edge tracing technique were also developed to determine the building area as the final step. The experiments show 88.9 % detection rate on the average and 95.9% for navigation video without trees and signs. The proposed algorithm can be also applied to point-of-interest (POI) detection for location based services (LBS) in ubiquitous computing environment.

Disposition of information entities and adequate level of information presentation in an in-car augmented reality navigation system

As Augmented Reality Navigation systems receive increasing attention as the next generation navigation systems, the importance of the layout of the elements and searching for the appropriate level of information cannot be overstated. Due to this, a series of experiments of usability tests had been designed. According to the experiments, the results for the screen arrangement showed that the MBN screen displayed on the left and the ARN screen on the right was preferred the most while the results for changing lanes showed that displaying information directly on the road and highlighting the path was the best method. As for switching direction, using the upper icons and displaying information directly on the road was preferred. The adequate amount of information was found out to be an average of 3 POI with 20pt font sizes for ARN displays and 10 POI with 18pt font sizes for MBN displays. Lastly, the Eye-tracking experiments showed that the ARN screen was viewed more often with a ratio of 7:3.

A Framework for the Development of Robust Lane Recognition Systems

Video-based navigation systems can provide more realistic navigation information to drivers by guiding their directions on a live video. To make video-based navigation systems, a robust lane recognition system is essential for 1) various road conditions such as damaged road and occluded road markings due to cars on the road, and 2) poor illumination conditions such as backlight, sunset, and so on. In this paper, a framework to make a robust lane recognition system is presented, which can overcome the difficulties mentioned above. In the proposed framework, a priori knowledge of the shape of road markings is exploited to reduce false positives and color information of lane markings are carefully analyzed to remove the poor illumination conditions. Based on the idea that the illumination effects to the background region, i.e., a set of pixels in the road region without lane markings, can be used to eliminate the illumination effects to lane markings, the proposed approach is designed to distinguish yellow lane markings from white lane markings. Experimental results are provided to show the effectiveness of the proposed lane recognition system for a various road and illumination conditions.