Ukil Yang

Gradient-Enhancing Conversion for Illumination-Robust Lane Detection

Lane detection is important in many advanced driver-assistance systems (ADAS). Vision-based lane detection algorithms are widely used and generally use gradient information as a lane feature. However, gradient values between lanes and roads vary with illumination change, which degrades the performance of lane detection systems. In this paper, we propose a gradient-enhancing conversion method for illumination-robust lane detection. Our proposed gradient-enhancing conversion method produces a new gray-level image from an RGB color image based on linear discriminant analysis. The converted images have large gradients at lane boundaries. To deal with illumination changes, the gray-level conversion vector is dynamically updated. In addition, we propose a novel lane detection algorithm, which uses the proposed conversion method, adaptive Canny edge detector, Hough transform, and curve model fitting method. We performed several experiments in various illumination environments and confirmed that the gradient is maximized at lane boundaries on the road. The detection rate of the proposed lane detection algorithm averages 96% and is greater than 93% in very poor environments.

Local feature based 3d face recognition

This paper presents a 3D face recognition system based on geometrically localized facial features. We propose the feature extraction procedure using the geometrical characteristics of a face. We extract three curvatures, eight invariant facial feature points and their relative features. These features are directly applied to face recognition algorithms which are a depth-based DP (Dynamic Programming) and a feature-based SVM (Support Vector Machine). Experimental results show that face recognition rates based on the depth-based DP and the feature-based SVM are 95% for 20 people and 96% for 100 people, respectively.

Illumination-invariant color space and its application to skin-color detection

Color-based digital image processing (DIP) techniques have attracted much attention in many vision-based applications. However, due to color variations resulting from illumination changes, many color-based DIP techniques have yet to demonstrate a stable state of performance. Skin-color detection, which is one of the popular color-based DIP techniques, must overcome the illumination problems. We address the issue by presenting an illumination-invariant color space based on the image acquisition model that is determined by the Lambertian surface. Furthermore, we propose a method of skin-color detection based on the illumination-invariant color space. To evaluate the performance in terms of the illumination-invariant property, we perform a skin-color detection experiment. In the experiment, we compare the proposed method with the methods based on several color spaces. From the experiment, we achieve encouraging results, and our empirical experiments evidence both the effectiveness and the usefulness of the proposed method.

3D face recognition based on facial shape indexes with dynamic programming

This paper describes a 3D face recognition method using facial shape indexes. Given an unknown range image, we extract invariant facial features based on the facial geometry. We estimate the 3D head pose using the proposed error compensated SVD method. For face recognition method, we define and extract facial shape indexes based on facial curvature characteristics and perform dynamic programming. Experimental results show that the proposed method is capable of determining the angle of faces accurately over a wide range of poses. In addition, 96.8% face recognition rate has been achieved based on the proposed method with 300 individuals with seven different poses.

Rear obstacle detection system with fisheye stereo camera using HCT

Abstract A vision based rear obstacle detection system is one of the most essential technologies, which can be used in many applications such as a parking assistance systems and intelligent vehicles. Stereo matching based obstacle detection methods have two inherent limitations, including sensitivity to illumination variation and high computational complexity. To overcome these problems, we present a hierarchical census transform (HCT)-based stereo matching method, and proposes a real-time rear obstacle detection system using fisheye stereo cameras. It includes two major components: firstly, we use a hierarchical approach to improve computational efficiency and reduce dependency on a matching window size. Computation time can also be accelerated with compute unified device architecture (CUDA) implementation because we designed the proposed method for parallel processing. Secondly, our method provides more accurate depth information, especially for moving objects, through cost aggregation using color and motion information. Experimental results show satisfactory performance under various real parking environments. The performance of the proposed method is superior to those of the conventional methods in terms of runtime and accuracy of depth estimation. Moreover, correct detection rate of the proposed system is 12.51% better than those of other systems, while its false and miss detection rates are 11.09% and 12.51% lower than those of other detection systems.

3D Face Recognition under Pose Varying Environments

This paper describes a novel three-dimensional (3D) face recognition method when the head pose varies severely. Given an unknown 3D face, we extract several invariant facial features based on the facial geometry. We perform a Error Compensated Singular Value Decomposition (EC-SVD) for 3D face recognition. The novelty of the proposed EC-SVD procedure lies in compensating for the error for each rotation axis accurately. When the pose of a face is estimated, we propose a novel two-stage 3D face recognition algorithm. We first select face candidates based on the 3D-based nearest neighbor classifier and then the depth-based template matching is performed for final recognition. From the experimental results, less than a 0.2 degree error in average has been achieved for the 3D head pose estimation and all faces are correctly matched based on our proposed method.

Illumination invariant skin color segmentation

Skin color segmentation takes a great attention in many vision-based methodologies. However, the performance of the segmentation is not stable because of color variations caused by various illumination conditions. In this paper, we propose a new skin color model for segmentation which is invariant to illumination variation. It is based on illumination-free color space which is defined on Lambertian surface. Through several experimental results, we confirm that the proposed skin color model is successfully applied to skin color segmentation under various illumination conditions.

An illumination invariant skin-color model for face detection

Face detection is an important step towards a fully automatic face recognition system. Among existing techniques in the literature, methods based on skin-color information have shown computational effectiveness as well as robustness in terms of rotation, scaling and partial occlusion. However, due to color variations resulted from illumination changes, many color-based techniques have yet to demonstrate a stable state of performance. In this paper, we present an illumination invariant color space model to address the color variation issue. The proposed method is evaluated both in terms of skin-color detection and face detection performances. Our empirical experiments evidenced both effectiveness and usefulness of the proposed method.

Pose and illumination invariant 2D to 3D facial recognition system

This paper proposes a pose and illumination invariant face recognition method based on a 2D to 3D facial recognition system which uses two dimensional (2D) image as an input and three dimensional (3D) data as a database. To improve the performance of a facial recognition system, we reorganize the framework of a conventional recognition system into more suitable framework for a 2D to 3D facial recognition system. 2D to 3D pose and illumination estimation algorithm is proposed based on a learning algorithm using multilayer perceptron (MLP). The proposed method estimates both pose and illumination factors of an input image in real-time, and compensates for a database in order to overcome the problems of pose and illumination without any occlusion occurred by insufficient information. To evaluate the performance, we performed both the accuracy tests of pose and illumination estimation and the recognition tests of 2D to 3D facial recognition system with a face database containing both two and three dimensional data.

Image Color Registration for Illumination-Robust Pattern Recognition

Image-based pattern recognition techniques have attracted much attention in vision-based applications. Color-based methods have shown several benefits. However, due to color variations resulting from illumination changes, many color-based techniques have yet to demonstrate stable performance. For illumination-robust pattern recognition, we propose an image color registration method based on an image acquisition model. Since the image acquisition model is created using the variables related to an illumination condition, camera characteristics, and an object’s surface reflectance, the proposed method normalizes the image’s color by taking into account both the illumination and camera characteristics. To evaluate the performance of the proposed method in terms of illumination-robust pattern recognition, we perform both an image similarity test and a feature similarity test between images acquired under different illumination conditions. Through the experiments, the superiority and the usefulness of the proposed method was validated.