Kyekyung Kim

Learning-based approach for license plate recognition

Presents a learning-based approach for the construction of a license-plate recognition system. The system consists of three modules. They are, respectively, the car detection module, the license-plate segmentation module and the recognition module. The car detection module detects a car in a given image sequence obtained from a camera with a simple color-based approach. The segmentation module extracts the license plate in the detected car image using neural networks as filters for analyzing the color and texture properties of the license plate. The recognition module then reads the characters on the detected license plate with a support vector machine (SVM)-based character recognizer. The system has been tested with 1000 video sequences obtained from toll-gates, parking lots, etc., and has shown the following performances on average: car detection rate 100%, segmentation rate 97.5%, and character recognition rate about 97.2%.

Character recognition system for cellular phone with camera

This paper describes camera based character recognition system, which is implemented for mobile devices such as PDA and cellular phones with color cameras. First, we have developed camera based character recognition system for PC that includes techniques such as image enhancement, local adaptive binarization and blob coloring to effectively extract character region and remove the noise of camera captured images. Then, we converted the PC based OCR system to the embedded OCR system for cellular phones. Several functions were specially developed since most of mobile telecommunication devices have no real number computing functions and have limitation of memory space. In this paper, we are addressing the related problems and our system mechanisms.

Object recognition for cell manufacturing system

The development of cell manufacturing process using object recognition has been interested in automated factory. But it is not trivial work to recognize object because features transformed from illumination and diversified field needs have caused challenge problem in object detection and recognition. The recognition reliability in real world environment can be increased by object, which preserves inherent feature and has invariance feature to scale, rotation or translation. In this paper, an illumination and rotation invariant object recognition is proposed. First, a binary image reserving clean object edges is achieved using DoG filter and local adaptive binarization. An object region from background is extracted with compensated edges that reserves geometry information of object. The object is recognized using neural network, which is trained with object classes that are categorized by object type and rotation angle. Standard shape model represented object class is used to estimate the pose of recognized object, which is handled by a robot. The simulation has been processed to evaluate feasibility of the proposed method that shows the accuracy of 99.86% and the matching speed of 0.03 seconds on ETRI database, which has 16,848 object images that has captured in various lighting environment.

Bin picking for the objects of non-Lambertian reflectance without using an explicit object model

Objects with non-Lambertian reflecting surfaces suffer severe appearance changes according to the change of observers angle of view. Thus it is difficult to obtain a consistent object model. This paper addresses the problem of bin picking for such objects without using explicit object models. Due to the above-mentioned reasons, traditional pattern matching approaches are difficult to apply. Furthermore, 3D sensors often fail to capture these objects correctly, which also disables the approaches based on 3D object models. Considering these limitations, this paper proposes a novel algorithm that tries to find a reasonable pickup point by investigating the status of the pile in the bin instead of trying to perfectly isolate single object from the pile and measure full pose information. For the purpose, feature point matching between stereo cameras is adopted to retrieve approximate 3D information of the pile. Additionally, by associating these corresponding point pairs with the binary patches around the point, corresponding binary patch pairs are detected, which is evaluated by a proposed objective function to select the most appropriate candidate for the pickup operation.

Distortion corrections for better character recognition of camera-based document images

The usage of cellular camera phones and digital cameras is rapidly increasing, but camera imaging applications are not so much due to the lack of practical camera imaging technology. Especially the acquisition environments of camera images are very different from those of scanner images. Light illumination, viewing distance and viewing angles constantly varies when we take a picture in indoor and outdoor. These variations make it difficult to extract character areas from images through binarization and the variation of camera viewing angles makes the images distorted geometrically. In this paper, these problems are totally discussed and the resolving methods are suggested for a better image recognition. The solutions such as adaptive binarization, color conversion, correction of lens distortion and correction of geometrical distortion are discussed and the sequence of correction processes are suggested for accurate document image recognition. In experiment, we use the various types of document images captured by mobile phone cameras and digital cameras. The results of distortion correction show that our image processing methods are efficient to increase the accuracy of character recognition for camera based document image.

Vision-based bin picking system for industrial robotics applications

The vision-based bin picking using object recognition has been considered as an innovative manufacturing process in industrial robotics applications. In bin picking system, pick and place tasks are performed by robot which has been processed by measuring object pose. But it has to address challenge problems such as object appearance distorted by overlapping parts, lighting variation or reflection, picking from randomly piled parts in a bin. This research is to investigate a vision-based bin-picking method, which provides a robust and efficient method to recognize object and to estimate pose with multiple vision sensors.

Rider posture analysis for postural correction on a sports simulator

For the use of sports simulators, there is a need for the implementation of proper functions for the realistic movement mechanism, the relative riding content, and the effective personalized training. In this paper, we present a posture analysis system for a rider on a sports simulator. The proposed system provides a personalized postural training function suited to the user based on their logging information and postural correction data. The system has an analysis function for the riding postures of the user based on the detected posture information. The user can see the error position with the riding contents and be corrected their posture in real time. We also set up a simulator test environment and performed experiments on the proposed system.

Precise object detection using local feature for robot manipulator

It is difficult to apply object recognition technology to manufacturing industry because the intrinsic characteristics of an object is easily influenced by the surrounding environment such as lighting condition, background complexity and object shape. This paper proposes a precise object detection method for assembling components by more stable feature extraction. To accomplish it, two images are captured by fine-tuning exposure time for the purpose of complementation of features. Adaptive binarization and differential of Gaussian methods are applied to extracting edge information from each image. In the next step, primary features such as contour lines are extracted from the object using the Fast Hough Transform and candidate lines are selected to become geometric conditions of the object. The precise object region is detected by shape analysis using the four vertices computed by the candidate lines. In addition, the internal features of the object are employed to increase the precision of object detection. As a result, the proposed method improved the accuracy of object detection so that it can be useful in the visual servoing using the robot manipulator.

Precise angle estimation using geometry features for bin picking

In the recent automation process, the manufacturing automation innovation is being done by using vision technology that can make instant judgement. In this paper, a method of vision based angle estimation is presented. The central moment is calculated and the rotation angle is estimated by the angle with the principal axis. There is also a method of estimating the precision rotation angle using the angle between matching points of SIFT descriptors. We propose a rotation angle estimation method that is invariant to object posture. We used affine transformation for the preprocessing of the image for the invariance to object posture. To estimate the accuracy of the rotation angle, we used blobs which are features in the object area. We conducted the angle estimation accuracy and repeatability tests under the experimental environment which has been constructed in our lab. We obtain less than 0.154 angle estimation accuracy.

Development of a postural analysis system for dinghy yacht simulators

In this paper, we present a postural analysis system for sailing user on dinghy yacht simulators. The proposed system detects the partitioned specific body parts of the user, extracts feature points of the body parts, and estimates posture information of the user based on the feature points. The extracted feature points are used as information for determining the position and posture of the sailing user. The system also analyzes and evaluates the applicable position and posture with the current situation information for the yachting environment. We set up a dinghy yacht simulator test environment and performed experiments on the proposed system.