Kwae-Hi Lee

Parking space detection using ultrasonic sensor in parking assistance system

This paper deals with parking space detection by using ultrasonic sensor. Using the multiple echo function, the accuracy of edge detection was increased. After inspecting effect on the multiple echo function in indoor experiment, we applied to 11 types of vehicles in real parking environment and made experiments on edge detection with various values of resolution. We can scan parking space more accurately in real parking environment. We propose the diagonal sensor to get information about the side of parking space. Our proposed method has benefit calculation and implementation is very simple.

Rain Removal Using Kalman Filter in Video

In outdoor environment vision system, weather effects are complex noise. They degrade the performance of vision algorithm such as feature detection, segmentation and object recognition. Rain scene has property that an image pixel is never always covered by rain throughout the whole video. Even though we can only see a short distance ahead because of heavy rain, each pixel of the scenes taken by camera is not always covered by rain. If pixel is affected by rain, this pixel has large intensity but this pixel has original chromatic information. Consequently we make a recovery to original color value via decreasing intensity to original value. Proposing recursive data processing, we can estimate the intensity value of each pixel. And this method needs the periodic reset. If unsuitable estimated mean value is determined because of heavy noise, henceforth cleanly captured scene has little weight effect for estimated mean value. Accordingly we reset the weighted value between estimated intensity and new captured one. Additionally, the proposed method has big advantages, including small equation by repeatedly estimating the mean of intensity and handling in real-time and benefits that too simple and it can implement to real time processing. And we obtain more pure scene than existing methods.

Visibility Enhancement Based Real -- Time Retinex for Diverse Environments

This paper proposes a modified structure of Retinex algorithm for the image enhancement method. The proposed algorithm performs robustly and stably on various environments, particularly in bad weather and night vision. Especially, Retinex algorithm tends to show outstanding results as compared with other approaches, and also shows advantages at night vision. However, the Retinex algorithm has problems such as the halo effect and the color distortion. In this paper, the proposed algorithm has a simple idea to remove above problems that combines a reflectance and a luminance image with some constant factors respectively. Since the result includes the luminance component, it works on diverse environment such as not only foggy image but also night vision image. Furthermore, this algorithm has proper structure for real time processing. Results on a variety of images show the effectiveness of this approach.

Estimation of motion blur parameters using cepstrum analysis

In this paper, we propose a novel estimation method of the parameters of motion blur, which is caused by the camera or object motion during the image and video capture. In the proposed method, the motion blur is characterized by the blur orientation and length in cepstrum domain. To a better estimation of orientation and length of the motion blur in the cepstrum domain, we use the difference of Gaussian as the input of the cepstrum transform, without additional transform such as Hough or Radon transform. We propose a simple and effective method, which separates blur components from image components. Simulations with various sets of simulated and real motion blurred images show the effectiveness of the proposed algorithm in terms of the quality. The proposed algorithm is computationally efficient and easy to use.

Localization Method Using Vector-Histogram

In this paper, we propose the localization algorithm using vector histogram matching method in environments with perpendicular walls. Conventional histogram matching method gets the wrong rotated value, because it has matched between other angle values when is a sharp rotation. In our proposed method, we do match angle values using direction of data. This method gets reference data and current data from the laser range finder. It calculates a magnitude and direction of a point adjacent. It makes a direction histogram from the scan data and calculates a rotated value of the mobile robot using cross correlation. The current data compensate as a rotated value. After getting the main direction from the reference data, two scan data make parallel to x-axis based on the main direction. Then using two scan data make x-axis histogram and y-axis histogram. It gets Deltaalpha,Deltabeta using cross correlation, where Deltaalpha and Deltabeta are variations of x-axis and y-axis. Translation value Deltax, Deltay calculates using rotated value and Deltaalpha, Deltabeta . Consequently, it accomplishes a self-localization of the mobile robot through problem solving of the histogram matching method.

A hierarchical stereo matching algorithm using wavelet representation based on edge and area information

A hybrid approach that uses edge information is proposed. At the first step, edge pixels are matched using adaptive windows that vary their shapes according to the direction of edge. At the next step, pixels outside edges are matched using adaptive windows that are limited by edge contours. The authors also constructed a pyramid structure using a wavelet transform. The disparity information obtained by matching low resolution images is utilized for interpolating the disparities of high resolution pixels unmatched.

An efficient motion adaptive de-interlacing algorithm using spatial and temporal filter

A motion adaptive de-interlacing algorithm is widely used for de-interlacing technique. Generally, a spatial method with edge dependent interpolation (EDI) algorithm shows better results than any other de-interlacing algorithms using single field in motion region and a temporal filter which is using multi field shows good results in stationary region. The motion adaptive method uses both a spatial and a temporal method. However, it also has problems with applying in high frequency image due to some artifacts. In this paper, we propose a de-interlacing algorithm with a motion adaptive method using EDI and temporal filter. The algorithm consists of an improved motion detection, EDI-based spatial filter, and vertical temporal filter (VTF). This algorithm produces a good visual performances in the case of fast moving objects and complex textures. PSNR shows

Digital elevation model combination using triangular image warping interpolation and maximum likelihood

The main purpose of this letter was to construct a large-region Digital Elevation Model (DEM) from small-area DEMs. Image mosaicking and data fusion techniques are necessary in constructing a large-region DEM from small-area DEMs. The topographic distortion of the image must be corrected because of viewing angle and time differences among the acquired images. The orthorectification and image mosaicking are executed using a grid matrix of the longitude and latitude obtained from DEM extraction at 10 m intervals. The height at each grid location in the overlapping region is determined using a triangular image warping interpolation and a maximum likelihood procedure. The images used in the experiments are two pairs of images from the SPOT High Resolution Visible (HRV) sensor. The results show that a large-region DEM can be obtained from small-area DEMs.

Automated reconstruction of urban house roofs from aerial imagery

In this paper, an approach for the automated reconstruction of urban house roofs from aerial images is proposed. We model a house roof as a set of planar polygonal patches, each of which is extracted by object related image segmentation and coplanar grouping. For object related Image segmentation we employ a region segmentation method by the distance transform. Coplanar grouping and polygonal patch formation is performed by selecting 3D line segments that are matched across different views using epipolar geometry and flight information. The algorithm has been applied to aerial images and the results show accurate reconstruction of urban house roofs.

A experiment of 3D reconstruction using laser range finder and CCD camera

This paper describe a experiment of 3D-Reconstruction using Laser range finder and CCD camera. In this experiment, we used a system that is composed of Senor part and Data processing part. This system is loading on the vehicle. We use two laser range finder, the one is horizontal direction scanner and the other is vertical direction scanner. Using Horizontal scanner can acquire the self-localization through objects(buildings etc.). And the vertical scanner can acquire information of object’s depth. When did experiment on long distance, This self-localization method has a cumulative error. So, in this paper, we will solve the solution. And then we remove obstacles of 3D-reconstructed building. Texture information is acquired with ccd camera. This texture data unite with range data, is created the 3D-virtual reconstruction of buildings. These techniques apply to city plan, 3D-environment game, movie background, unmanned-patrol etc.