Minkyu Cheon

Vision-Based Vehicle Detection System With Consideration of the Detecting Location

In this paper, we propose a vision-based vehicle detection system. We use a method composed of a hypothesis generation (HG) step and a hypothesis verification (HV) step, following the general approach to vision-based vehicle detection systems. In the HG step, the system extracts hypotheses using shadow regions that appear under vehicles. In the HV step, the system classifies feature vectors extracted from hypotheses to determine whether those hypotheses are vehicles. Along with the histogram of oriented gradients (HOG), we propose and implement a new type of feature vector, i.e., HOG symmetry vectors, in this paper. We also propose a new classification method that uses data importance in the HV step. The data importance value is based on the locations of hypotheses to prioritize hypotheses that have greater risks of accident. Experimental results show the strong performance of our proposed system.

Object tracking from image sequences using adaptive models in fuzzy particle filter

This paper describes a vision-based system for tracking objects from image sequences. The proposed system has the standard architecture with a particle filter which is a popular algorithm to track objects in real time. Many tracking algorithms have a great difficulty in tracking objects robustly by reason of complex background and rapid changes under a real complex environment such as a traffic road. To make a robust algorithm for object tracking, we propose the method that uses the adaptive autoregressive model as a state transition model and the adaptive appearance mixture model as an observation model. But, in case of changing the state of a tracked object suddenly, the adaptive models may not make the optimal parameters for accurate states at current time. Because the noise variance of the adaptive models in this case is larger than that in normal case, it has an effect on the accuracy of an object tracking algorithm. Thus, we propose a fuzzy particle filter to overcome problems from the occurrence of the unexpected improper variances due to several causes. In this paper, as the process noises and the observation noises in a fuzzy particle filter are considered as fuzzy variables by using the possibility theory, a fuzzy particle filter with fuzzy noises is used to manage uncertainty in various noise models. Also, we make possibility measure as using the fuzzy relation equation which is defined by these fuzzy variables. And then, the states are estimated by using a fuzzy expected value operator. Also, because the proposed algorithm applies several functions to improve the accuracy of tracking an object, the performance of tracking speed deteriorates. To resolve this problem to some extent, we consider the fact that a fuzzy particle filter has a little bit of an effect on the number of particles. Consequently, we propose the method which can adjust the number of particles by using the result from a measurement step in order to improve the speed for an object tracking in the proposed algorithm. The experiments of this paper show that the proposed method is efficient and has many advantages for an object tracking in real environments.

Adaptive background model for non-static background subtraction by estimation of the color change ratio

Background modeling, a preliminary processing step for foreground detection, is a challenging task because of the complexity and variety of background regions and unexpected scenarios such as sudden illumination changes, waving trees, rippling water, etc. In this work, we develop a pixel-based background modeling method that uses a probabilistic approach by means of changing color sequences. This method uses two background models in tandem. The first model uses a static background, which is obtained via a probabilistic approach and is a standard from which the foreground is extracted. The second method uses an adaptive background, which is modeled by the degree of color change. This background functions as an additional standard from which the foreground is extracted and is appropriate for eliminating non-static background elements. These models enable the developed method to automatically adapt to various environments. The algorithm was tested on various video sequences and its performance was evaluated by comparison with other state-of-the-art background subtraction methods.

Best Basis Selection Method Using Learning Weights for Face Recognition

In the face recognition field, principal component analysis is essential to the reduction of the image dimension. In spite of frequent use of this analysis, it is commonly believed that the basis faces with large eigenvalues are chosen as the best subset in the nearest neighbor classifiers. We propose an alternative that can predict the classification error during the training steps and find the useful basis faces for the similarity metrics of the classical pattern algorithms. In addition, we also show the need for the eye-aligned dataset to have the pure face. The experiments using face images verify that our method reduces the negative effect on the misaligned face images and decreases the weights of the useful basis faces in order to improve the classification accuracy.

An Improvement of AdaBoost using Boundary Classifier

The method proposed in this paper can improve the performance of the Boosting algorithm in machine learning. The proposed Boundary AdaBoost algorithm can make up for the weak points of Normal binary classifier using threshold boundary concepts. The new proposed boundary can be located near the threshold of the binary classifier. The proposed algorithm improves classification in areas where Normal binary classifier is weak. Thus, the optimal boundary final classifier can decrease error rates classified with more reasonable features. Finally, this paper derives the new algorithm’s optimal solution, and it demonstrates how classifier accuracy can be improved using the proposed Boundary AdaBoost in a simulation experiment of pedestrian detection using 10-fold cross validation.

Intelligent emergency stop algorithm for a manipulator using a new regression method

In working environments with large manipulators, accidental collisions can cause severe personal injuries and can seriously damage manipulators, necessitating the development of an emergency stop algorithm to prevent such occurrences. In this paper, we propose an emergency stop system for the efficient and safe operation of a manipulator by applying an intelligent emergency stop algorithm. Our proposed intelligent algorithm considers the direction of motion of the manipulator. In addition, using a new regression method, the algorithm includes a decision step that determines whether a detected object is a collision-causing obstacle or a part of the manipulator. We apply our emergency stop system to a two-link manipulator and assess the performance of our intelligent emergency stop algorithm as compared with other models.

Distance Sensitive AdaBoost using Distance Weight Function

This paper proposes a new method to improve performance of AdaBoost by using a distance weight function to increase the accuracy of its machine learning processes. The proposed distance weight algorithm improves classification in areas where the original binary classifier is weak. This paper derives the new algorithm’s optimal solution, and it demonstrates how classifier accuracy can be improved using the proposed Distance Sensitive AdaBoost in a simulation experiment of pedestrian detection.

Rotation Invariant Histogram of Oriented Gradients

In this paper, we propose a new image descriptor, that is, a rotation invariant histogram of oriented gradients (RIHOG). RIHOG overcomes a disadvantage of the histogram of oriented gradients (HOG), which is very sensitive to image rotation. The HOG only uses magnitude values of a pixel without considering neighboring pixels. The RIHOG uses the accumulated relative magnitude values of corresponding relative orientation calculated with neighboring pixels, which has an effect on reducing the sensitivity to image rotation. The performance of RIHOG is verified via the index of classification and classification of Brodatz texture data.

Development of Fuzzy Support Vector Machine and Evaluation of Performance Using Ionosphere Radar Data

Support Vector machine is the classifier which is based on the statistical training theory. Twin Support Vector Machine(TWSVM) is a kind of binary classifier that determines two nonparallel planes by solving two related SVM-type problems. The training time of TWSVM is shorter than that of SVM, but TWSVM doesn`t shows worse performance than that of SVM. This paper proposes the TWSVM which is applied fuzzy membership, and compares the performance of this classifier with the other classifiers using Ionosphere radar data set.