In recent years, safety technologies of cars are becoming important and the cases that cameras are mounted in cars are increasing. Since the perception of the distance of a usual image is difficult for drivers, method to construct overhead view images and assist drivers are proposed. In this paper, we propose methods for estimating camera parameters of multiple cameras and construct overhead view images with small errors. Effectiveness of the methods is examined by experiments.

View full abstract

A root cause detection method for semiconductor fabrication process based on analysis of a defect distribution pattern of a wafer (wafer map) was developed. We show two methods of root cause analysis. One method is the comparison analysis that determines a tool of the root cause by searching a tool monitoring wafers having similar wafer map to a product wafer. The other method is the commonality analysis that finds the tool commonly processing product wafers having similar wafer maps. To search similar wafer map automatically, a comparison algorithm of two sets of distributed points was developed. This algorithm measures a similarity of two wafer maps based on density images whose pixel value shows corresponding defect density. The two methods including the developed algorithm were evaluated using the data obtained from a real semiconductor fabrication process. The accuracy of the comparison analysis method was 91.7% for 12 cases in which the tool of the root cause were detected, while false rate was 0% for 22 cases in which there are no similar patterns in tool monitoring wafers. For 6 cases that the tool of root cause were detected by the commonality analysis, the algorithm classified wafers into similar patterns and non-similar patterns correctly. These results show that the proposed methods are applicable to the semiconductor fabrication process.

View full abstract

It is difficult to detect simultaneously both the local defects such as scratch and the global defects such as obscure dirt with sufficient sensitivity by image processing. In order to solve such a problem, a method for detecting globally distributed defects by using learning with Mahalanobis distance is proposed in this paper. First, n-dimensional hierarchical average processing is performed in a image, and the n-dimensional feature space is parallelly defined corresponding to all pixels of the image. Through this procedure, the information with more global circumstances could be included in each pixel as the dimension goes up. Continuously, if the training sample image is learned by using this n-dimensional feature space, the super-boundary surface given by the equal discriminant with the Mahalanobis distance in n-dimensional feature space of each pixel will become a super-ellipse. Then both the local defects such as scratch and global defects such as obscure dirt or stain are simultaneously detected with high sensitivity by using the discriminant super-boundary surface.

View full abstract

A slow-speed-notifying signal is a quite important component of railway signal system that is related with the train's safety. However, because detection of the temporary signal is depending on only the driver's eyes, and furthers a backup system such as ATS is not installed, the security of the train's safety is not complete. Therefore, in order to solve this problem, we have been developing a system based on image processing method, which uses pattern-matching algorithm. This system is able to detect slow-speed-notifying signals automatically from image sequences of front view of a train, and gives an alarm to the drivers to assist them. In this paper, we describe the details of these fundamental algorithms and the results of prototype program's benchmark tests.

View full abstract

A new surface profiling technique is proposed, which enables us to measure a surface profile without losing high frequency components of sharp edges and with extended phase measurement range. It is accomplished by a newly developed two-wavelength imaging system and the local model fitting (LMF) algorithm for carrier fringe pattern analysis. The experimental results demonstrated that the proposed method can measure steep steps fast and accurately with an extended measurement range, while the robustness against vibration in the conventional single-shot methods is maintained.

View full abstract

The purpose of this paper is to propose a new approach to obstacles detection using a single camera mounted on a vehicle when the vehicle is backing or turning round at an intersection at a low speed. Using equations among feature point locations and optical flows in geometrically transferred top-view images, ground-movement information can be estimated. We compensate for the ground-movement between consecutive top-view images using the estimated ground-movement information and compute the difference image using the compensated previous top-view image and the current top-view image. Finally, obstacle regions on top-view image can be easily extracted using the difference image. The proposed approach is demonstrated to be effective by evaluation images captured by an in-vehicle camera

View full abstract

The use of asynchronous multiple pan-tilt cameras makes it difficult to calculate 3D position of a moving object such as a ball in sports scene. This paper presents a method for calculating the 3D position using asynchronous pan-tilt cameras. The 3D positions are calculated from motion trajectory. Then, the 3D position of latest frame is calculated from previous 3D positions. The 3D position is recalculated using constraints from ray information. Since the 3D position is calculated at every shutter timing of each cameras, the 3D position can be obtained at number of camera × 30 points. We show that the proposed method achieves accurate 3D measurement using asynchronous pan-tilt cameras.

View full abstract

An imaging system was build using a telecentric optics in image space and a CCD image sensor that was movable along the optical axis in order to keep a perfect perspective projection during focusing. It was successfully applied to establish omni-directional measurement of a drill bit as an example of shiny metallic objects based on the shape-from-focus method. Negative data from glare regions were replaced with overlapping positive data obtained from neighboring viewpoints. Usage of a table that precisely rotates the object simplified registration of the multiple-view data. Finally, a fine mesh connecting surface points (corresponding pixels in images) can be generated to produce a three-dimensional graphic model with texture.

View full abstract

In recent years, accidents of factories especially nuclear power plants by human errors and natural disaster by tsunami and earthquake has been happened so many times. Therefore, much research on computer warning systems at these various disasters has been done. Actually, these damages are decreased by these warning system. So it can be said that these system are effective. These computer warning systems researched in the past judge a current state of human and systems by a human action as a result to warnings. Therefore it is considered that these systems will be better by estimation of a inner state of human on the way to the state after human action. In order to estimate inner state of human, an analysis of human expression is enumerated. And surprise emotion in human emotion is very primitive and is often occurred in dangerous situation. In considering of these characteristics of surprised expression, it is very effective for construction of computer warning systems to be able to analyze surprised expression not to shift panic state. In the first step of this research, we observed rough occurring structure of surprised expression. In the next step, we analyzed detailed movement of surprise expression taken by high-speed camera.

View full abstract

We present a method of registering a sequence of aerial consecutive images taken from an airborne sensor into a mosaic, as well as a method for visualizing moving objects on the large satellite image, thus providing information about traffic conditions following a disaster. Our approach uses features obtained from histograms of oriented gradients (HOG) and the scale invariant feature transform (SIFT) to register both the aerial images and a large satellite image. Since the SIFT is used to find keypoints for image registration and the HOG is used to localize these points by using local texture information, we registered a precise image sequence even though the images were not captured at the same time. We then applied moving object detection on the basis of the results of tracking the points from the registered images and visualized them on the satellite image. We present the results on real image sequences from a helicopter, and show that visualization of moving objects is a useful way to display road conditions over a wide area in a disaster situation.

View full abstract

The components of the food related to the “deliciousness” are usually evaluated by the componential analysis. Component content and the kind of the component in the food are measured by this analysis. However, componential analysis is not able to analyze detailed measurement, and it needs long time to measure. We propose a method to measure two-dimensional distribution of the component in the food by using the near infrared ray (IR) image. An advantage of our method is to be able to make a visualized image of invisible components, because the component content is measured by taking subtraction of two wavelength of near IR light. Many components of the food have characteristics in the range of IR. In this paper, we describe a method to measure the component of the food by using near IR image processing, and we show an application to visualize the saccharose in the pumpkin.

View full abstract

This paper addresses an approach to robust search based on color images observed in condition of fluctuation in illumination spectrum. A new search method, named 'Color Orientation Code Matching' is proposed, which introduces weighting into the conventional Orientation Code Matching based on monochrome images. The weighting is based on evaluation of stationarity in spectrum in order to avoid mis-matching, under ill-conditions. The experimental results with actual data demonstrate the effectiveness of robust images searching in fluctuation of color spectrum.

View full abstract

This study paid attention to the disagreement of the eye-contact in teleconference caused by the separation between the input camera and output monitor devices. Then the authors proposed the Eye-Contact Camera System for generating eye-contacted motion images to the receiver. In this system, iris contour is extracted after the face region extraction, the vertical and horizontal directions of the glance are calculated based on the relation among positions of the monitor, camera and receiver, and finally the iris center coordinates are shifted in the image so that the partner looks just looking at him, and vice versa. This paper suggested that improved iris detection for coping with roll rotation of face, and implemented the system on note PC with web-camera for evaluating the usability. As a result, iris detection rate was improved from 32% to 65%, and usability More than 92% (25/27) users gave positive evaluation to eye-contacted image.

View full abstract