Dong-Joong Kang

Real-time lane detection by using multiple cues

People have a growing interest for driver assistant systems that are used to monitor the driving conditions by visual technique, and warn and guide drivers the road conditions. This paper proposes a real-time lane detection algorithm which is a necessary part for driver assistant system and unmanned vehicle. The algorithm presented in this paper integrates multiple cues, including bar filter which is efficient to detect bar-shape objects like road lane, color cue, and Hough Transform (HT). After obtaining integrated multiple cues we utilize particle filtering technique to realize lane tracking, which guarantees the robust and real-time lane detection. Experimental results show that the algorithm gives a precise and robust detection of lane in various situations.

Specimen alignment in an axial tensile test of thin films using direct imaging and its influence on the mechanical properties of BeCu

This paper proposes a new system for verification of the alignment of loading fixtures and test specimens during tensile testing of thin film with a micrometer size through direct imaging. The novel and reliable image recognition system to evaluate the misalignment between the load train and the specimen axes during tensile test of thin film was developed using digital image processing technology with CCD. The decision of whether alignment of the tensile specimen is acceptable or not is based on a probabilistic analysis through the edge feature extraction of digital imaging. In order to verify the performance of the proposed system and investigate the effect of the misalignment of the specimen on tensile properties, the tensile tests were performed as displacement control in air and at room temperature for metal thin film, the beryllium copper (BeCu) alloys. In the case of the metal thin films, bending stresses caused by misalignment are insignificant because the films are easily bent during tensile tests to eliminate the bending stresses. And it was observed that little effects and scatters on tensile properties occur by stress gradient caused by twisting at in-plane misalignment, and the effects and scatters on tensile properties are insignificant at out-of-plane misalignment, in the case of the BeCu thin film.

New measurement method of Poisson's ratio of PVA hydrogels using an optical flow analysis for a digital imaging system

In this paper, we propose a method to observe deformation, to measure strain and to calculate Poissons ratio of a soft material such as a polyvinyl alcohol hydrogel by means of an optical flow analysis during a tensile test. Compared to the conventional digital image correlation method of which the basic assumption is that the local target region in an image is under linear deformation, an optical flow method can measure local deformation up to the pixel level of the image. In addition, sub-pixel measurements are possible by means of bi-cubic interpolation of each flow vector. To guarantee the accuracy of the optical flow vectors, the scale-invariant feature transform was used. Random sample consensus (RANSAC) for eliminating noisy features and then obtaining a more accurate result was applied. The mean value of Poissons ratio is 0.4498 and the standard deviation comes to 0.0305. This optical flow method can also be applied to calculate Poissons ratio of other soft materials.

New strain measurement method at axial tensile test of thin films through direct imaging

This paper proposes a new method for measuring strain during a tensile test of the specimen with micrometre size through direct imaging. A specimen was newly designed for adoption of direct imaging which was the main contribution of the proposed system. The structure of the specimen has eight indicators that make it possible to adopt direct imaging and it is fabricated using the same process of microelectromechanical system (MEMS) devices to guarantee the feasibility of the tensile test. We implemented a system for non-contact in situ measurement of strain with the specimen, the image-based displacement measurement system. Extension of the gauge length in the specimen could be found robustly by computing the positions of the eight rectangular-shape indicators on the image. Also, for an easy setup procedure, the region of interest was found automatically through the analysis of the edge projection profile along the horizontal direction. To gain confidence in the reliability of the system, the tensile test for the Al–3%Ti thin film was performed, which is widely used as a material in MEMS devices. Tensile tests were performed and displacements were measured using the proposed method and also the capacitance type displacement sensor for comparison. It is demonstrated that the new strain measurement system can be effectively used in the tensile test of the specimen at microscale with easy setup and better accuracy.

Ellipse detection: a simple and precise method based on randomized Hough transform

We present a simple ellipse detector that accurately extracts the parameters of an ellipse based on randomized Hough transform (RHT). Ellipse detection by the conventional RHT method is challenging due to the huge calculation burden and voting complexity for the five parameters of one ellipse. To address this, we extracted formulas that separated these five parameters into two to three parameters and proposed a separated two-level voting scheme based on the RHT. The original image was first processed by edge detection, eight-zone distribution of its direction, and edge lists merging, and then the parameters were calculated and voted by the separated two-level voting scheme. Finally, an evaluation method was used to determine whether or not the detected ellipse existed in the image. We tested our method on various kinds of real images, and the experiments demonstrated that the proposed method provided a precise and efficient ellipse detection.

Easy calibration method of vision system for in-situ measurement of strain of thin films

An easy calibration method was presented for in-situ measurement of displacement in the order of nanometer during micro-tensile test for thin films by using CCD camera as a sensing device. The calibration of the sensing camera in the system is a central element part to measure displacement in the order of nanometer using images taken with the camera. This was accomplished by modeling the optical projection through the camera lens and relative locations between the object and camera in 3D space. A set of known 3D points on a plane where the film is located on is projected to an image plane as input data. These points, known as a calibration points, are then used to estimate the projection parameters of the camera. In the measurement system of the micro-scale by CCD camera, the calibration data acquisition and one-to-one matching steps between the image and 3D planes need precise data extraction procedures and repetitive users operation to calibrate the measuring devices. The lack of the robust image feature extraction and easy matching prevent the practical use of these methods. A data selection method was proposed to overcome these limitations and offer an easy and convenient calibration of a vision system that has the CCD camera and the 3D reference plane with calibration marks of circular type on the surface of the plane. The method minimizes the users intervention such as the fine tuning of illumination system and provides an efficient calibration method of the vision system for in-situ axial displacement measurement of the micro-tensile materials.

A Method to Improve the Performance of Adaboost Algorithm by Using Mixed Weak Classifier

The weak classifier of AdaBoost algorithm is a central classification element that uses a single criterion separating positive and negative learning candidates. Finding the best criterion to separate two feature distributions influences learning capacity of the algorithm. A common way to classify the distributions is to use the mean value of the features. However, positive and negative distributions of Haar-like feature as an image descriptor are hard to classify by a single threshold. The poor classification ability of the single threshold also increases the number of boosting operations, and finally results in a poor classifier. This paper proposes a weak classifier that uses multiple criterions by adding a probabilistic criterion of the positive candidate distribution with the conventional mean classifier: the positive distribution has low variation and the values are closer to the mean while the negative distribution has large variation and values are widely spread. The difference in the variance for the positive and negative distributions is used as an additional criterion. In the learning procedure, we use a new classifier that provides a better classifier between them by selective switching between the mean and standard deviation. We call this new type of combined classifier the “Mixed Weak Classifier”. The proposed weak classifier is more robust than the mean classifier alone and decreases the number of boosting operations to be converged.

Method to improve the performance of the adaboost algorithm by combining weak classifiers

The weak classifier of AdaBoost algorithm is a central classification element that uses a single criterion separating positive and negative learning candidates. Finding the best criterion to separate two feature distributions influences learning capacity of the algorithm. A common way to classify the distributions is to use the mean value of the features. However, positive and negative distributions of Haar-like feature as an image descriptor are hard to classify by a single threshold. The poor classification ability of the single threshold also increases the number of boosting operations, and finally results in a poor classifier. This paper proposes a weak classifier that uses multiple criterions by adding the standard deviation (STD) of the positive candidate distribution with the conventional mean classifier: the positive distribution has low variation and the values are closer to the mean while the negative distribution has large variation and values are widely spread. The difference in the STD for the positive and negative distributions is used as an additional criterion. In the learning procedure, we use a new classifier that provides a better classifier between them by selective switching between the mean and standard deviation. We call this new type of combined classifier the ldquoMixed Weak Classifierrdquo. The proposed weak classifier is more robust than the mean classifier alone and decreases the number of boosting operations to be converged.

Ellipse detection using an improved randomized Hough transformation

This paper proposes an ellipse detection algorithm based on the analytical solution to the parameters of ellipse in images. In the first instance, edge detection is processed, from which line segments are extracted. Then the method of finding the center coordinates of the ellipse is described based on the property of ellipse by using three points voting at a sense of randomized Hough Transformation (RHT). Finally, an analytical solution of the other three parameters of the ellipse (semi-major axis length, semi-minor axis length and the angle between the X-axis and the major axis of the ellipse) are given via coordinate transformation. Based on this solution, we propose the separated parameter voting scheme for ellipse center and the other three parameters instead of 5 parameters voting scheme of RHT. The experiments show that the proposed algorithm performs well in various images.

Learning-based human detection using difference image in visual surveillance

In visual surveillance, accurate detection of each human is important for various application of counting and tracking of people. Applying general human detection algorithm for each image could be applied. In this paper, we propose a learning-based human segmentation algorithm. Histogram of Oriented Gradient (HOG) shows remarkable result on human detection and it uses intensity image. We use difference image as a training sample and it is obtained through the accumulation of multiple difference images. We show that proposed algorithm could be a good candidate for the fast generation of possible regions of human in visual surveillance. We show the feasibility of proposed algorithm using publicly available data sets.